diff --git a/.devops/cpu.Dockerfile b/.devops/cpu.Dockerfile index 8d020f16c..522ee8147 100644 --- a/.devops/cpu.Dockerfile +++ b/.devops/cpu.Dockerfile @@ -2,6 +2,10 @@ ARG UBUNTU_VERSION=22.04 FROM ubuntu:$UBUNTU_VERSION AS build +ARG TARGETARCH + +ARG GGML_CPU_ARM_ARCH=armv8-a + RUN apt-get update && \ apt-get install -y build-essential git cmake libcurl4-openssl-dev @@ -9,7 +13,14 @@ WORKDIR /app COPY . . -RUN cmake -S . -B build -DGGML_BACKEND_DL=ON -DGGML_NATIVE=OFF -DGGML_CPU_ALL_VARIANTS=ON -DLLAMA_CURL=ON -DCMAKE_BUILD_TYPE=Release && \ +RUN if [ "$TARGETARCH" = "amd64" ]; then \ + cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DLLAMA_CURL=ON -DGGML_NATIVE=OFF -DGGML_BACKEND_DL=ON -DGGML_CPU_ALL_VARIANTS=ON; \ + elif [ "$TARGETARCH" = "arm64" ]; then \ + cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DLLAMA_CURL=ON -DGGML_NATIVE=OFF -DGGML_CPU_ARM_ARCH=${GGML_CPU_ARM_ARCH}; \ + else \ + echo "Unsupported architecture"; \ + exit 1; \ + fi && \ cmake --build build -j $(nproc) RUN mkdir -p /app/lib && \ diff --git a/.github/ISSUE_TEMPLATE/010-bug-compilation.yml b/.github/ISSUE_TEMPLATE/010-bug-compilation.yml index f10b3a2b2..b85bf5741 100644 --- a/.github/ISSUE_TEMPLATE/010-bug-compilation.yml +++ b/.github/ISSUE_TEMPLATE/010-bug-compilation.yml @@ -65,12 +65,22 @@ body: If possible, please do a git bisect and identify the exact commit that introduced the bug. validations: required: false + - type: textarea + id: command + attributes: + label: Compile command + description: > + Please provide the exact command you used to compile llama.cpp. For example: `cmake -B ...`. + This will be automatically formatted into code, so no need for backticks. + render: shell + validations: + required: true - type: textarea id: logs attributes: label: Relevant log output description: > - Please copy and paste any relevant log output, including the command that you entered and any generated text. + Please copy and paste any relevant log output, including any generated text. This will be automatically formatted into code, so no need for backticks. render: shell validations: diff --git a/.github/ISSUE_TEMPLATE/019-bug-misc.yml b/.github/ISSUE_TEMPLATE/019-bug-misc.yml index d157ea307..1904e31fd 100644 --- a/.github/ISSUE_TEMPLATE/019-bug-misc.yml +++ b/.github/ISSUE_TEMPLATE/019-bug-misc.yml @@ -52,6 +52,16 @@ body: - Other (Please specify in the next section) validations: required: false + - type: textarea + id: command + attributes: + label: Command line + description: > + Please provide the exact commands you entered, if applicable. For example: `llama-server -m ... -c ...`, `llama-cli -m ...`, etc. + This will be automatically formatted into code, so no need for backticks. + render: shell + validations: + required: false - type: textarea id: info attributes: @@ -74,7 +84,7 @@ body: attributes: label: Relevant log output description: > - If applicable, please copy and paste any relevant log output, including the command that you entered and any generated text. + If applicable, please copy and paste any relevant log output, including any generated text. This will be automatically formatted into code, so no need for backticks. render: shell validations: diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml index 2128eab4f..4f0663c0e 100644 --- a/.github/workflows/build.yml +++ b/.github/workflows/build.yml @@ -56,6 +56,7 @@ jobs: mkdir build cd build cmake .. \ + -DCMAKE_BUILD_RPATH="@loader_path" \ -DLLAMA_FATAL_WARNINGS=ON \ -DLLAMA_CURL=ON \ -DGGML_METAL_USE_BF16=ON \ @@ -87,6 +88,7 @@ jobs: if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }} run: | cp LICENSE ./build/bin/ + cp examples/run/linenoise.cpp/LICENSE ./build/bin/LICENSE.linenoise.cpp zip -r llama-${{ steps.tag.outputs.name }}-bin-macos-arm64.zip ./build/bin/* - name: Upload artifacts @@ -119,6 +121,7 @@ jobs: # Metal is disabled due to intermittent failures with Github runners not having a GPU: # https://github.com/ggerganov/llama.cpp/actions/runs/8635935781/job/23674807267#step:5:2313 cmake -B build \ + -DCMAKE_BUILD_RPATH="@loader_path" \ -DLLAMA_FATAL_WARNINGS=ON \ -DLLAMA_CURL=ON \ -DGGML_METAL=OFF \ @@ -149,6 +152,7 @@ jobs: if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }} run: | cp LICENSE ./build/bin/ + cp examples/run/linenoise.cpp/LICENSE ./build/bin/LICENSE.linenoise.cpp zip -r llama-${{ steps.tag.outputs.name }}-bin-macos-x64.zip ./build/bin/* - name: Upload artifacts @@ -158,8 +162,8 @@ jobs: path: llama-${{ steps.tag.outputs.name }}-bin-macos-x64.zip name: llama-bin-macos-x64.zip - ubuntu-latest-cmake: - runs-on: ubuntu-latest + ubuntu-cpu-cmake: + runs-on: ubuntu-22.04 steps: - name: Clone @@ -179,7 +183,10 @@ jobs: run: | mkdir build cd build - cmake .. -DLLAMA_FATAL_WARNINGS=ON -DLLAMA_CURL=ON -DGGML_RPC=ON + cmake .. \ + -DLLAMA_FATAL_WARNINGS=ON \ + -DLLAMA_CURL=ON \ + -DGGML_RPC=ON cmake --build . --config Release -j $(nproc) - name: Test @@ -217,6 +224,7 @@ jobs: if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }} run: | cp LICENSE ./build/bin/ + cp examples/run/linenoise.cpp/LICENSE ./build/bin/LICENSE.linenoise.cpp zip -r llama-${{ steps.tag.outputs.name }}-bin-ubuntu-x64.zip ./build/bin/* - name: Upload artifacts @@ -234,7 +242,7 @@ jobs: strategy: matrix: sanitizer: [ADDRESS, THREAD, UNDEFINED] - build_type: [Debug, Release] + build_type: [Debug] steps: - name: Clone @@ -253,7 +261,10 @@ jobs: run: | mkdir build cd build - cmake .. -DLLAMA_FATAL_WARNINGS=ON -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} + cmake .. \ + -DLLAMA_FATAL_WARNINGS=ON \ + -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON \ + -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} cmake --build . --config ${{ matrix.build_type }} -j $(nproc) - name: Build (no OpenMP) @@ -262,7 +273,11 @@ jobs: run: | mkdir build cd build - cmake .. -DLLAMA_FATAL_WARNINGS=ON -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} -DGGML_OPENMP=OFF + cmake .. \ + -DLLAMA_FATAL_WARNINGS=ON \ + -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON \ + -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} \ + -DGGML_OPENMP=OFF cmake --build . --config ${{ matrix.build_type }} -j $(nproc) - name: Test @@ -292,7 +307,8 @@ jobs: run: | mkdir build cd build - cmake -DGGML_RPC=ON .. + cmake .. \ + -DGGML_RPC=ON cmake --build . --config Release -j $(nproc) - name: Test @@ -322,7 +338,8 @@ jobs: run: | mkdir build cd build - cmake -DGGML_VULKAN=ON .. + cmake .. \ + -DGGML_VULKAN=ON cmake --build . --config Release -j $(nproc) - name: Test @@ -349,13 +366,18 @@ jobs: - name: Build with native CMake HIP support id: cmake_build run: | - cmake -B build -S . -DCMAKE_HIP_COMPILER="$(hipconfig -l)/clang" -DGGML_HIP=ON + cmake -B build -S . \ + -DCMAKE_HIP_COMPILER="$(hipconfig -l)/clang" \ + -DGGML_HIP=ON cmake --build build --config Release -j $(nproc) - name: Build with legacy HIP support id: cmake_build_legacy_hip run: | - cmake -B build2 -S . -DCMAKE_C_COMPILER=hipcc -DCMAKE_CXX_COMPILER=hipcc -DGGML_HIP=ON + cmake -B build2 -S . \ + -DCMAKE_C_COMPILER=hipcc \ + -DCMAKE_CXX_COMPILER=hipcc \ + -DGGML_HIP=ON cmake --build build2 --config Release -j $(nproc) ubuntu-22-cmake-musa: @@ -376,7 +398,8 @@ jobs: - name: Build with native CMake MUSA support id: cmake_build run: | - cmake -B build -S . -DGGML_MUSA=ON + cmake -B build -S . \ + -DGGML_MUSA=ON cmake --build build --config Release -j $(nproc) ubuntu-22-cmake-sycl: @@ -417,7 +440,10 @@ jobs: source /opt/intel/oneapi/setvars.sh mkdir build cd build - cmake -DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx .. + cmake .. \ + -DGGML_SYCL=ON \ + -DCMAKE_C_COMPILER=icx \ + -DCMAKE_CXX_COMPILER=icpx cmake --build . --config Release -j $(nproc) ubuntu-22-cmake-sycl-fp16: @@ -458,42 +484,13 @@ jobs: source /opt/intel/oneapi/setvars.sh mkdir build cd build - cmake -DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DGGML_SYCL_F16=ON .. + cmake .. \ + -DGGML_SYCL=ON \ + -DCMAKE_C_COMPILER=icx \ + -DCMAKE_CXX_COMPILER=icpx \ + -DGGML_SYCL_F16=ON cmake --build . --config Release -j $(nproc) - # TODO: build with GGML_METAL=OFF because test-backend-ops fail on "Apple Paravirtual device" and I don't know - # how to debug it. - # ref: https://github.com/ggerganov/llama.cpp/actions/runs/7132125951/job/19422043567?pr=4359#step:5:6584 - # would be great if we fix these - macOS-latest-cmake: - runs-on: macos-latest - - steps: - - name: Clone - id: checkout - uses: actions/checkout@v4 - - - name: Dependencies - id: depends - continue-on-error: true - run: | - brew update - - - name: Build - id: cmake_build - run: | - sysctl -a - mkdir build - cd build - cmake -DLLAMA_FATAL_WARNINGS=ON -DGGML_METAL=OFF .. - cmake --build . --config Release -j $(sysctl -n hw.logicalcpu) - - - name: Test - id: cmake_test - run: | - cd build - ctest -L main --verbose --timeout 900 - macOS-latest-cmake-ios: runs-on: macos-latest @@ -665,7 +662,7 @@ jobs: - build: 'llvm-arm64' defines: '-G "Ninja Multi-Config" -D CMAKE_TOOLCHAIN_FILE=cmake/arm64-windows-llvm.cmake -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON' - build: 'msvc-arm64' - defines: '-G "Ninja Multi-Config" -D CMAKE_TOOLCHAIN_FILE=cmake/arm64-windows-msvc.cmake -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DBUILD_SHARED_LIBS=O' + defines: '-G "Ninja Multi-Config" -D CMAKE_TOOLCHAIN_FILE=cmake/arm64-windows-msvc.cmake -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON' - build: 'llvm-arm64-opencl-adreno' defines: '-G "Ninja Multi-Config" -D CMAKE_TOOLCHAIN_FILE=cmake/arm64-windows-llvm.cmake -DCMAKE_PREFIX_PATH="$env:RUNNER_TEMP/opencl-arm64-release" -DGGML_OPENCL=ON -DGGML_OPENCL_USE_ADRENO_KERNELS=ON' @@ -796,6 +793,7 @@ jobs: if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }} run: | Copy-Item LICENSE .\build\bin\Release\llama.cpp.txt + Copy-Item .\examples\run\linenoise.cpp\LICENSE .\build\bin\Release\linenoise.cpp.txt 7z a llama-${{ steps.tag.outputs.name }}-bin-win-${{ matrix.build }}.zip .\build\bin\Release\* - name: Upload artifacts @@ -823,7 +821,13 @@ jobs: - name: Build with CMake run: | - cmake -S . -B build -G Ninja -DCMAKE_BUILD_TYPE=Release -DGGML_NATIVE=OFF -DGGML_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES=89-real -DCMAKE_EXE_LINKER_FLAGS=-Wl,--allow-shlib-undefined -DLLAMA_FATAL_WARNINGS=ON + cmake -S . -B build -G Ninja \ + -DCMAKE_BUILD_TYPE=Release \ + -DCMAKE_CUDA_ARCHITECTURES=89-real \ + -DCMAKE_EXE_LINKER_FLAGS=-Wl,--allow-shlib-undefined \ + -DLLAMA_FATAL_WARNINGS=ON \ + -DGGML_NATIVE=OFF \ + -DGGML_CUDA=ON cmake --build build windows-2019-cmake-cuda: @@ -912,7 +916,11 @@ jobs: shell: cmd run: | call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat" - cmake -S . -B build -G "Ninja Multi-Config" -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DGGML_RPC=ON + cmake -S . -B build -G "Ninja Multi-Config" ^ + -DLLAMA_BUILD_SERVER=ON ^ + -DGGML_NATIVE=OFF ^ + -DGGML_CUDA=ON ^ + -DGGML_RPC=ON set /A NINJA_JOBS=%NUMBER_OF_PROCESSORS%-1 cmake --build build --config Release -j %NINJA_JOBS% -t ggml cmake --build build --config Release @@ -1065,7 +1073,12 @@ jobs: run: | $env:HIP_PATH=$(Resolve-Path 'C:\Program Files\AMD\ROCm\*\bin\clang.exe' | split-path | split-path) $env:CMAKE_PREFIX_PATH="${env:HIP_PATH}" - cmake -G "Unix Makefiles" -B build -S . -DCMAKE_C_COMPILER="${env:HIP_PATH}\bin\clang.exe" -DCMAKE_CXX_COMPILER="${env:HIP_PATH}\bin\clang++.exe" -DGGML_HIP=ON -DCMAKE_BUILD_TYPE=Release -DGGML_RPC=ON + cmake -G "Unix Makefiles" -B build -S . ` + -DCMAKE_C_COMPILER="${env:HIP_PATH}\bin\clang.exe" ` + -DCMAKE_CXX_COMPILER="${env:HIP_PATH}\bin\clang++.exe" ` + -DCMAKE_BUILD_TYPE=Release ` + -DGGML_HIP=ON ` + -DGGML_RPC=ON cmake --build build -j ${env:NUMBER_OF_PROCESSORS} windows-latest-cmake-hip-release: @@ -1103,7 +1116,13 @@ jobs: run: | $env:HIP_PATH=$(Resolve-Path 'C:\Program Files\AMD\ROCm\*\bin\clang.exe' | split-path | split-path) $env:CMAKE_PREFIX_PATH="${env:HIP_PATH}" - cmake -G "Unix Makefiles" -B build -S . -DCMAKE_C_COMPILER="${env:HIP_PATH}\bin\clang.exe" -DCMAKE_CXX_COMPILER="${env:HIP_PATH}\bin\clang++.exe" -DGGML_HIP=ON -DCMAKE_BUILD_TYPE=Release -DAMDGPU_TARGETS=${{ matrix.gpu_target }} -DGGML_RPC=ON + cmake -G "Unix Makefiles" -B build -S . ` + -DCMAKE_C_COMPILER="${env:HIP_PATH}\bin\clang.exe" ` + -DCMAKE_CXX_COMPILER="${env:HIP_PATH}\bin\clang++.exe" ` + -DCMAKE_BUILD_TYPE=Release ` + -DAMDGPU_TARGETS=${{ matrix.gpu_target }} ` + -DGGML_HIP=ON ` + -DGGML_RPC=ON cmake --build build -j ${env:NUMBER_OF_PROCESSORS} cp "${env:HIP_PATH}\bin\hipblas.dll" "build\bin\" @@ -1194,8 +1213,7 @@ jobs: runs-on: ubuntu-latest needs: - - ubuntu-latest-cmake - - macOS-latest-cmake + - ubuntu-cpu-cmake - windows-latest-cmake - windows-2019-cmake-cuda - windows-latest-cmake-hip-release @@ -1234,7 +1252,7 @@ jobs: - name: Create release id: create_release - uses: anzz1/action-create-release@v1 + uses: ggml-org/action-create-release@v1 env: GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }} with: @@ -1454,3 +1472,37 @@ jobs: # popd # emcmake cmake . -DCMAKE_BUILD_TYPE=${{ matrix.build }} # make + + openEuler-latest-cmake-cann: + if: ${{ github.event_name != 'pull_request' || contains(github.event.pull_request.labels.*.name, 'Ascend NPU') }} + defaults: + run: + shell: bash -el {0} + runs-on: ubuntu-24.04-arm + strategy: + matrix: + cann: + - '8.0.rc3.beta1-910b-openeuler22.03-py3.10' + device: + - 'ascend910b3' + build: + - 'Release' + container: ascendai/cann:${{ matrix.cann }} + steps: + - name: Checkout + uses: actions/checkout@v4 + + - name: Dependencies + run: | + yum update -y + yum install -y git gcc gcc-c++ make cmake + + - name: Build + run: | + export LD_LIBRARY_PATH=${ASCEND_TOOLKIT_HOME}/lib64:${ASCEND_TOOLKIT_HOME}/$(uname -m)-linux/devlib/:${LD_LIBRARY_PATH} + + cmake -S . -B build \ + -DCMAKE_BUILD_TYPE=${{ matrix.build }} \ + -DGGML_CANN=on \ + -DSOC_TYPE=${{ matrix.device }} + cmake --build build -j $(nproc) diff --git a/.github/workflows/docker.yml b/.github/workflows/docker.yml index 41f1a89ee..d71f1eb38 100644 --- a/.github/workflows/docker.yml +++ b/.github/workflows/docker.yml @@ -97,10 +97,9 @@ jobs: GITHUB_BRANCH_NAME: ${{ github.head_ref || github.ref_name }} GITHUB_REPOSITORY_OWNER: '${{ github.repository_owner }}' - # https://github.com/jlumbroso/free-disk-space/tree/54081f138730dfa15788a46383842cd2f914a1be#example - name: Free Disk Space (Ubuntu) if: ${{ matrix.config.free_disk_space == true }} - uses: jlumbroso/free-disk-space@main + uses: ggml-org/free-disk-space@v1.3.1 with: # this might remove tools that are actually needed, # if set to "true" but frees about 6 GB diff --git a/.github/workflows/editorconfig.yml b/.github/workflows/editorconfig.yml index ae86e9927..f02b7c219 100644 --- a/.github/workflows/editorconfig.yml +++ b/.github/workflows/editorconfig.yml @@ -23,5 +23,7 @@ jobs: runs-on: ubuntu-latest steps: - uses: actions/checkout@v4 - - uses: editorconfig-checker/action-editorconfig-checker@main + - uses: editorconfig-checker/action-editorconfig-checker@v2 + with: + version: v3.0.3 - run: editorconfig-checker diff --git a/.github/workflows/server.yml b/.github/workflows/server.yml index 671fe595c..ed1c357a5 100644 --- a/.github/workflows/server.yml +++ b/.github/workflows/server.yml @@ -112,9 +112,9 @@ jobs: -DGGML_OPENMP=OFF ; cmake --build build --config ${{ matrix.build_type }} -j $(nproc) --target llama-server - - name: Build - id: cmake_build - if: ${{ matrix.sanitizer != 'THREAD' }} + - name: Build (sanitizers) + id: cmake_build_sanitizers + if: ${{ matrix.sanitizer != '' && matrix.sanitizer != 'THREAD' }} run: | cmake -B build \ -DGGML_NATIVE=OFF \ @@ -124,12 +124,31 @@ jobs: -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON ; cmake --build build --config ${{ matrix.build_type }} -j $(nproc) --target llama-server + - name: Build (sanitizers) + id: cmake_build + if: ${{ matrix.sanitizer == '' }} + run: | + cmake -B build \ + -DGGML_NATIVE=OFF \ + -DLLAMA_BUILD_SERVER=ON \ + -DLLAMA_CURL=ON \ + -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} ; + cmake --build build --config ${{ matrix.build_type }} -j $(nproc) --target llama-server + - name: Tests id: server_integration_tests + if: ${{ matrix.sanitizer == '' }} run: | cd examples/server/tests ./tests.sh + - name: Tests (sanitizers) + id: server_integration_tests_sanitizers + if: ${{ matrix.sanitizer != '' }} + run: | + cd examples/server/tests + LLAMA_SANITIZE=1 ./tests.sh + - name: Slow tests id: server_integration_tests_slow if: ${{ (github.event.schedule || github.event.inputs.slow_tests == 'true') && matrix.build_type == 'Release' }} diff --git a/.gitignore b/.gitignore index 1df7cf4a1..694f36e04 100644 --- a/.gitignore +++ b/.gitignore @@ -18,6 +18,7 @@ *.metallib *.o *.so +*.swp *.tmp # IDE / OS diff --git a/CMakeLists.txt b/CMakeLists.txt index a717a508f..e7f520582 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -16,6 +16,7 @@ endif() list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/cmake/") set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/bin) +set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/bin) if (CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR) set(LLAMA_STANDALONE ON) @@ -49,6 +50,7 @@ endif() if (MSVC) add_compile_options("$<$:/utf-8>") add_compile_options("$<$:/utf-8>") + add_compile_options(/bigobj) endif() # @@ -83,11 +85,8 @@ include(${CMAKE_CURRENT_SOURCE_DIR}/cmake/build-info.cmake) include(${CMAKE_CURRENT_SOURCE_DIR}/cmake/common.cmake) # override ggml options -set(GGML_SANITIZE_THREAD ${LLAMA_SANITIZE_THREAD}) -set(GGML_SANITIZE_ADDRESS ${LLAMA_SANITIZE_ADDRESS}) -set(GGML_SANITIZE_UNDEFINED ${LLAMA_SANITIZE_UNDEFINED}) -set(GGML_ALL_WARNINGS ${LLAMA_ALL_WARNINGS}) -set(GGML_FATAL_WARNINGS ${LLAMA_FATAL_WARNINGS}) +set(GGML_ALL_WARNINGS ${LLAMA_ALL_WARNINGS}) +set(GGML_FATAL_WARNINGS ${LLAMA_FATAL_WARNINGS}) # change the default for these ggml options if (NOT DEFINED GGML_LLAMAFILE) @@ -117,16 +116,62 @@ llama_option_depr(WARNING LLAMA_SYCL GGML_SYCL) llama_option_depr(WARNING LLAMA_SYCL_F16 GGML_SYCL_F16) llama_option_depr(WARNING LLAMA_CANN GGML_CANN) +if (NOT MSVC) + if (LLAMA_SANITIZE_THREAD) + message(STATUS "Using -fsanitize=thread") + + add_compile_options(-fsanitize=thread) + link_libraries (-fsanitize=thread) + endif() + + if (LLAMA_SANITIZE_ADDRESS) + message(STATUS "Using -fsanitize=address") + + add_compile_options(-fsanitize=address -fno-omit-frame-pointer) + link_libraries (-fsanitize=address) + endif() + + if (LLAMA_SANITIZE_UNDEFINED) + message(STATUS "Using -fsanitize=undefined") + + add_compile_options(-fsanitize=undefined) + link_libraries (-fsanitize=undefined) + endif() +endif() + # -# build the library +# 3rd-party # if (NOT TARGET ggml) add_subdirectory(ggml) # ... otherwise assume ggml is added by a parent CMakeLists.txt endif() + +# +# build the library +# + add_subdirectory(src) +# +# utils, programs, examples and tests +# + +if (LLAMA_BUILD_COMMON) + add_subdirectory(common) +endif() + +if (LLAMA_BUILD_COMMON AND LLAMA_BUILD_TESTS AND NOT CMAKE_JS_VERSION) + include(CTest) + add_subdirectory(tests) +endif() + +if (LLAMA_BUILD_COMMON AND LLAMA_BUILD_EXAMPLES) + add_subdirectory(examples) + add_subdirectory(pocs) +endif() + # # install # @@ -200,21 +245,3 @@ configure_file(cmake/llama.pc.in install(FILES "${CMAKE_CURRENT_BINARY_DIR}/llama.pc" DESTINATION lib/pkgconfig) - -# -# utils, programs, examples and tests -# - -if (LLAMA_BUILD_COMMON) - add_subdirectory(common) -endif() - -if (LLAMA_BUILD_COMMON AND LLAMA_BUILD_TESTS AND NOT CMAKE_JS_VERSION) - include(CTest) - add_subdirectory(tests) -endif() - -if (LLAMA_BUILD_COMMON AND LLAMA_BUILD_EXAMPLES) - add_subdirectory(examples) - add_subdirectory(pocs) -endif() diff --git a/CODEOWNERS b/CODEOWNERS index adeba5395..72d594b46 100644 --- a/CODEOWNERS +++ b/CODEOWNERS @@ -1,5 +1,11 @@ # collaborators can optionally add themselves here to indicate their availability for reviewing related PRs /ci/ @ggerganov -/.devops/ @ngxson +/.devops/*.Dockerfile @ngxson /examples/server/ @ngxson +/ggml/src/ggml-cuda/fattn* @JohannesGaessler +/ggml/src/ggml-cuda/mmq.* @JohannesGaessler +/ggml/src/ggml-cuda/mmv.* @JohannesGaessler +/ggml/src/ggml-cuda/mmvq.* @JohannesGaessler +/ggml/src/ggml-opt.cpp @JohannesGaessler +/ggml/src/gguf.cpp @JohannesGaessler diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md index 5a85ec5d2..8d411982b 100644 --- a/CONTRIBUTING.md +++ b/CONTRIBUTING.md @@ -1,10 +1,10 @@ # Pull requests (for contributors) - Test your changes: - - Execute [the full CI locally on your machine](ci/README.md) before publishing - - Verify that the perplexity and the performance are not affected negatively by your changes (use `llama-perplexity` and `llama-bench`) - - If you modified the `ggml` source, run the `test-backend-ops` tool to check whether different backend implementations of the `ggml` operators produce consistent results (this requires access to at least two different `ggml` backends) - - If you modified a `ggml` operator or added a new one, add the corresponding test cases to `test-backend-ops` + - Execute [the full CI locally on your machine](ci/README.md) before publishing + - Verify that the perplexity and the performance are not affected negatively by your changes (use `llama-perplexity` and `llama-bench`) + - If you modified the `ggml` source, run the `test-backend-ops` tool to check whether different backend implementations of the `ggml` operators produce consistent results (this requires access to at least two different `ggml` backends) + - If you modified a `ggml` operator or added a new one, add the corresponding test cases to `test-backend-ops` - Consider allowing write access to your branch for faster reviews, as reviewers can push commits directly - If your PR becomes stale, don't hesitate to ping the maintainers in the comments @@ -20,14 +20,104 @@ - Avoid adding third-party dependencies, extra files, extra headers, etc. - Always consider cross-compatibility with other operating systems and architectures - Avoid fancy-looking modern STL constructs, use basic `for` loops, avoid templates, keep it simple -- There are no strict rules for the code style, but try to follow the patterns in the code (indentation, spaces, etc.). Vertical alignment makes things more readable and easier to batch edit +- Vertical alignment makes things more readable and easier to batch edit - Clean-up any trailing whitespaces, use 4 spaces for indentation, brackets on the same line, `void * ptr`, `int & a` -- Naming usually optimizes for common prefix (see https://github.com/ggerganov/ggml/pull/302#discussion_r1243240963) +- Use sized integer types such as `int32_t` in the public API, e.g. `size_t` may also be appropriate for allocation sizes or byte offsets +- Declare structs with `struct foo {}` instead of `typedef struct foo {} foo` + - In C++ code omit optional `struct` and `enum` keyword whenever they are not necessary + ```cpp + // OK + llama_context * ctx; + const llama_rope_type rope_type; + + // not OK + struct llama_context * ctx; + const enum llama_rope_type rope_type; + ``` + + _(NOTE: this guideline is yet to be applied to the `llama.cpp` codebase. New code should follow this guideline.)_ + +- Try to follow the existing patterns in the code (indentation, spaces, etc.). In case of doubt use `clang-format` to format the added code +- For anything not covered in the current guidelines, refer to the [C++ Core Guidelines](https://isocpp.github.io/CppCoreGuidelines/CppCoreGuidelines) - Tensors store data in row-major order. We refer to dimension 0 as columns, 1 as rows, 2 as matrices - Matrix multiplication is unconventional: [`C = ggml_mul_mat(ctx, A, B)`](https://github.com/ggerganov/llama.cpp/blob/880e352277fc017df4d5794f0c21c44e1eae2b84/ggml.h#L1058-L1064) means $C^T = A B^T \Leftrightarrow C = B A^T.$ ![matmul](media/matmul.png) +# Naming guidelines + +- Use `snake_case` for function, variable and type names +- Naming usually optimizes for longest common prefix (see https://github.com/ggerganov/ggml/pull/302#discussion_r1243240963) + + ```cpp + // not OK + int small_number; + int big_number; + + // OK + int number_small; + int number_big; + ``` + +- Enum values are always in upper case and prefixed with the enum name + + ```cpp + enum llama_vocab_type { + LLAMA_VOCAB_TYPE_NONE = 0, + LLAMA_VOCAB_TYPE_SPM = 1, + LLAMA_VOCAB_TYPE_BPE = 2, + LLAMA_VOCAB_TYPE_WPM = 3, + LLAMA_VOCAB_TYPE_UGM = 4, + LLAMA_VOCAB_TYPE_RWKV = 5, + }; + ``` + +- The general naming pattern is `_`, with `` being `_` + + ```cpp + llama_model_init(); // class: "llama_model", method: "init" + llama_sampler_chain_remove(); // class: "llama_sampler_chain", method: "remove" + llama_sampler_get_seed(); // class: "llama_sampler", method: "get_seed" + llama_set_embeddings(); // class: "llama_context", method: "set_embeddings" + llama_n_threads(); // class: "llama_context", method: "n_threads" + llama_adapter_lora_free(); // class: "llama_adapter_lora", method: "free" + ``` + + - The `get` `` can be omitted + - The `` can be omitted if not necessary + - The `_context` suffix of the `` is optional. Use it to disambiguate symbols when needed + - Use `init`/`free` for constructor/destructor `` + +- Use the `_t` suffix when a type is supposed to be opaque to the user - it's not relevant to them if it is a struct or anything else + + ```cpp + typedef struct llama_context * llama_context_t; + + enum llama_pooling_type llama_pooling_type(const llama_context_t ctx); + ``` + + _(NOTE: this guideline is yet to be applied to the `llama.cpp` codebase. New code should follow this guideline)_ + +- C/C++ filenames are all lowercase with dashes. Headers use the `.h` extension. Source files use the `.c` or `.cpp` extension +- Python filenames are all lowercase with underscores + +- _(TODO: abbreviations usage)_ + +# Preprocessor directives + +- _(TODO: add guidelines with examples and apply them to the codebase)_ + + ```cpp + #ifdef FOO + #endif // FOO + ``` + +# Documentation + +- Documentation is a community effort +- When you need to look into the source code to figure out how to use an API consider adding a short summary to the header file for future reference +- When you notice incorrect or outdated documentation, please update it + # Resources The Github issues, PRs and discussions contain a lot of information that can be useful to get familiar with the codebase. For convenience, some of the more important information is referenced from Github projects: diff --git a/Makefile b/Makefile index 435241ea8..1c9b92bf3 100644 --- a/Makefile +++ b/Makefile @@ -1361,7 +1361,9 @@ llama-server: \ examples/server/httplib.h \ examples/server/index.html.hpp \ examples/server/loading.html.hpp \ + common/chat-template.hpp \ common/json.hpp \ + common/minja.hpp \ $(OBJ_ALL) $(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<) $(CXX) $(CXXFLAGS) $(filter-out %.h %.hpp $<,$^) -Iexamples/server $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS) $(LWINSOCK2) diff --git a/README.md b/README.md index 0126da89c..97d028670 100644 --- a/README.md +++ b/README.md @@ -16,7 +16,9 @@ Inference of Meta's [LLaMA](https://arxiv.org/abs/2302.13971) model (and others) ## Hot topics -- **Introducing GGUF-my-LoRA** https://github.com/ggerganov/llama.cpp/discussions/10123 +- **VS Code extension for FIM completions:** https://github.com/ggml-org/llama.vscode +- Vim/Neovim plugin for FIM completions: https://github.com/ggml-org/llama.vim +- Introducing GGUF-my-LoRA https://github.com/ggerganov/llama.cpp/discussions/10123 - Hugging Face Inference Endpoints now support GGUF out of the box! https://github.com/ggerganov/llama.cpp/discussions/9669 - Hugging Face GGUF editor: [discussion](https://github.com/ggerganov/llama.cpp/discussions/9268) | [tool](https://huggingface.co/spaces/CISCai/gguf-editor) @@ -69,6 +71,7 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo - [x] [Qwen models](https://huggingface.co/models?search=Qwen/Qwen) - [x] [PLaMo-13B](https://github.com/ggerganov/llama.cpp/pull/3557) - [x] [Phi models](https://huggingface.co/models?search=microsoft/phi) +- [x] [PhiMoE](https://github.com/ggerganov/llama.cpp/pull/11003) - [x] [GPT-2](https://huggingface.co/gpt2) - [x] [Orion 14B](https://github.com/ggerganov/llama.cpp/pull/5118) - [x] [InternLM2](https://huggingface.co/models?search=internlm2) @@ -98,6 +101,7 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo - [x] [Jais](https://huggingface.co/inceptionai/jais-13b-chat) - [x] [Bielik-11B-v2.3](https://huggingface.co/collections/speakleash/bielik-11b-v23-66ee813238d9b526a072408a) - [x] [RWKV-6](https://github.com/BlinkDL/RWKV-LM) +- [x] [QRWKV-6](https://huggingface.co/recursal/QRWKV6-32B-Instruct-Preview-v0.1) - [x] [GigaChat-20B-A3B](https://huggingface.co/ai-sage/GigaChat-20B-A3B-instruct) #### Multimodal @@ -202,6 +206,7 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo - [GPUStack](https://github.com/gpustack/gpustack) - Manage GPU clusters for running LLMs - [llama_cpp_canister](https://github.com/onicai/llama_cpp_canister) - llama.cpp as a smart contract on the Internet Computer, using WebAssembly - [llama-swap](https://github.com/mostlygeek/llama-swap) - transparent proxy that adds automatic model switching with llama-server +- [Kalavai](https://github.com/kalavai-net/kalavai-client) - Crowdsource end to end LLM deployment at any scale @@ -243,6 +248,8 @@ The [Hugging Face](https://huggingface.co) platform hosts a [number of LLMs](htt - [Trending](https://huggingface.co/models?library=gguf&sort=trending) - [LLaMA](https://huggingface.co/models?sort=trending&search=llama+gguf) +You can either manually download the GGUF file or directly use any `llama.cpp`-compatible models from Hugging Face by using this CLI argument: `-hf /[:quant]` + After downloading a model, use the CLI tools to run it locally - see below. `llama.cpp` requires the model to be stored in the [GGUF](https://github.com/ggerganov/ggml/blob/master/docs/gguf.md) file format. Models in other data formats can be converted to GGUF using the `convert_*.py` Python scripts in this repo. @@ -261,21 +268,12 @@ To learn more about model quantization, [read this documentation](examples/quant #### A CLI tool for accessing and experimenting with most of `llama.cpp`'s functionality. -
- Run simple text completion - - ```bash - llama-cli -m model.gguf -p "I believe the meaning of life is" -n 128 - - # I believe the meaning of life is to find your own truth and to live in accordance with it. For me, this means being true to myself and following my passions, even if they don't align with societal expectations. I think that's what I love about yoga – it's not just a physical practice, but a spiritual one too. It's about connecting with yourself, listening to your inner voice, and honoring your own unique journey. - ``` - -
- --
Run in conversation mode + Models with a built-in chat template will automatically activate conversation mode. If this doesn't occur, you can manually enable it by adding `-cnv` and specifying a suitable chat template with `--chat-template NAME` + ```bash - llama-cli -m model.gguf -p "You are a helpful assistant" -cnv + llama-cli -m model.gguf # > hi, who are you? # Hi there! I'm your helpful assistant! I'm an AI-powered chatbot designed to assist and provide information to users like you. I'm here to help answer your questions, provide guidance, and offer support on a wide range of topics. I'm a friendly and knowledgeable AI, and I'm always happy to help with anything you need. What's on your mind, and how can I assist you today? @@ -287,17 +285,28 @@ To learn more about model quantization, [read this documentation](examples/quant
-
- Run with custom chat template + Run in conversation mode with custom chat template ```bash - # use the "chatml" template - llama-cli -m model.gguf -p "You are a helpful assistant" -cnv --chat-template chatml + # use the "chatml" template (use -h to see the list of supported templates) + llama-cli -m model.gguf -cnv --chat-template chatml # use a custom template - llama-cli -m model.gguf -p "You are a helpful assistant" -cnv --in-prefix 'User: ' --reverse-prompt 'User:' + llama-cli -m model.gguf -cnv --in-prefix 'User: ' --reverse-prompt 'User:' ``` - [Supported templates](https://github.com/ggerganov/llama.cpp/wiki/Templates-supported-by-llama_chat_apply_template) +
+ +-
+ Run simple text completion + + To disable conversation mode explicitly, use `-no-cnv` + + ```bash + llama-cli -m model.gguf -p "I believe the meaning of life is" -n 128 -no-cnv + + # I believe the meaning of life is to find your own truth and to live in accordance with it. For me, this means being true to myself and following my passions, even if they don't align with societal expectations. I think that's what I love about yoga – it's not just a physical practice, but a spiritual one too. It's about connecting with yourself, listening to your inner voice, and honoring your own unique journey. + ```
diff --git a/ci/run.sh b/ci/run.sh index abf08a4ff..77c32ce00 100755 --- a/ci/run.sh +++ b/ci/run.sh @@ -326,17 +326,17 @@ function gg_run_open_llama_7b_v2 { ./bin/llama-quantize ${model_f16} ${model_q5_k} q5_k ./bin/llama-quantize ${model_f16} ${model_q6_k} q6_k - (time ./bin/llama-cli --model ${model_f16} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-f16.log - (time ./bin/llama-cli --model ${model_q8_0} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q8_0.log - (time ./bin/llama-cli --model ${model_q4_0} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_0.log - (time ./bin/llama-cli --model ${model_q4_1} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_1.log - (time ./bin/llama-cli --model ${model_q5_0} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_0.log - (time ./bin/llama-cli --model ${model_q5_1} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_1.log - (time ./bin/llama-cli --model ${model_q2_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q2_k.log - (time ./bin/llama-cli --model ${model_q3_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q3_k.log - (time ./bin/llama-cli --model ${model_q4_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_k.log - (time ./bin/llama-cli --model ${model_q5_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_k.log - (time ./bin/llama-cli --model ${model_q6_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q6_k.log + (time ./bin/llama-cli -no-cnv --model ${model_f16} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-f16.log + (time ./bin/llama-cli -no-cnv --model ${model_q8_0} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q8_0.log + (time ./bin/llama-cli -no-cnv --model ${model_q4_0} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_0.log + (time ./bin/llama-cli -no-cnv --model ${model_q4_1} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_1.log + (time ./bin/llama-cli -no-cnv --model ${model_q5_0} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_0.log + (time ./bin/llama-cli -no-cnv --model ${model_q5_1} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_1.log + (time ./bin/llama-cli -no-cnv --model ${model_q2_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q2_k.log + (time ./bin/llama-cli -no-cnv --model ${model_q3_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q3_k.log + (time ./bin/llama-cli -no-cnv --model ${model_q4_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_k.log + (time ./bin/llama-cli -no-cnv --model ${model_q5_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_k.log + (time ./bin/llama-cli -no-cnv --model ${model_q6_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q6_k.log (time ./bin/llama-perplexity --model ${model_f16} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-tg-f16.log (time ./bin/llama-perplexity --model ${model_q8_0} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-tg-q8_0.log @@ -460,17 +460,17 @@ function gg_run_pythia_1_4b { ./bin/llama-quantize ${model_f16} ${model_q5_k} q5_k ./bin/llama-quantize ${model_f16} ${model_q6_k} q6_k - (time ./bin/llama-cli --model ${model_f16} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-f16.log - (time ./bin/llama-cli --model ${model_q8_0} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q8_0.log - (time ./bin/llama-cli --model ${model_q4_0} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_0.log - (time ./bin/llama-cli --model ${model_q4_1} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_1.log - (time ./bin/llama-cli --model ${model_q5_0} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_0.log - (time ./bin/llama-cli --model ${model_q5_1} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_1.log - (time ./bin/llama-cli --model ${model_q2_k} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q2_k.log - (time ./bin/llama-cli --model ${model_q3_k} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q3_k.log - (time ./bin/llama-cli --model ${model_q4_k} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_k.log - (time ./bin/llama-cli --model ${model_q5_k} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_k.log - (time ./bin/llama-cli --model ${model_q6_k} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q6_k.log + (time ./bin/llama-cli -no-cnv --model ${model_f16} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-f16.log + (time ./bin/llama-cli -no-cnv --model ${model_q8_0} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q8_0.log + (time ./bin/llama-cli -no-cnv --model ${model_q4_0} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_0.log + (time ./bin/llama-cli -no-cnv --model ${model_q4_1} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_1.log + (time ./bin/llama-cli -no-cnv --model ${model_q5_0} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_0.log + (time ./bin/llama-cli -no-cnv --model ${model_q5_1} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_1.log + (time ./bin/llama-cli -no-cnv --model ${model_q2_k} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q2_k.log + (time ./bin/llama-cli -no-cnv --model ${model_q3_k} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q3_k.log + (time ./bin/llama-cli -no-cnv --model ${model_q4_k} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_k.log + (time ./bin/llama-cli -no-cnv --model ${model_q5_k} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_k.log + (time ./bin/llama-cli -no-cnv --model ${model_q6_k} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q6_k.log (time ./bin/llama-perplexity --model ${model_f16} -f ${wiki_test_60} -ngl 99 -c 128 -b 128 --chunks 1 ) 2>&1 | tee -a $OUT/${ci}-tg-f16.log (time ./bin/llama-perplexity --model ${model_q8_0} -f ${wiki_test_60} -ngl 99 -c 128 -b 128 --chunks 1 ) 2>&1 | tee -a $OUT/${ci}-tg-q8_0.log @@ -591,17 +591,17 @@ function gg_run_pythia_2_8b { ./bin/llama-quantize ${model_f16} ${model_q5_k} q5_k ./bin/llama-quantize ${model_f16} ${model_q6_k} q6_k - (time ./bin/llama-cli --model ${model_f16} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-f16.log - (time ./bin/llama-cli --model ${model_q8_0} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q8_0.log - (time ./bin/llama-cli --model ${model_q4_0} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_0.log - (time ./bin/llama-cli --model ${model_q4_1} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_1.log - (time ./bin/llama-cli --model ${model_q5_0} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_0.log - (time ./bin/llama-cli --model ${model_q5_1} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_1.log - (time ./bin/llama-cli --model ${model_q2_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q2_k.log - (time ./bin/llama-cli --model ${model_q3_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q3_k.log - (time ./bin/llama-cli --model ${model_q4_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_k.log - (time ./bin/llama-cli --model ${model_q5_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_k.log - (time ./bin/llama-cli --model ${model_q6_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q6_k.log + (time ./bin/llama-cli -no-cnv --model ${model_f16} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-f16.log + (time ./bin/llama-cli -no-cnv --model ${model_q8_0} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q8_0.log + (time ./bin/llama-cli -no-cnv --model ${model_q4_0} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_0.log + (time ./bin/llama-cli -no-cnv --model ${model_q4_1} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_1.log + (time ./bin/llama-cli -no-cnv --model ${model_q5_0} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_0.log + (time ./bin/llama-cli -no-cnv --model ${model_q5_1} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_1.log + (time ./bin/llama-cli -no-cnv --model ${model_q2_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q2_k.log + (time ./bin/llama-cli -no-cnv --model ${model_q3_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q3_k.log + (time ./bin/llama-cli -no-cnv --model ${model_q4_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_k.log + (time ./bin/llama-cli -no-cnv --model ${model_q5_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_k.log + (time ./bin/llama-cli -no-cnv --model ${model_q6_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q6_k.log (time ./bin/llama-perplexity --model ${model_f16} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-tg-f16.log (time ./bin/llama-perplexity --model ${model_q8_0} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-tg-q8_0.log diff --git a/cmake/build-info.cmake b/cmake/build-info.cmake index ea3dc55c8..c1a456e17 100644 --- a/cmake/build-info.cmake +++ b/cmake/build-info.cmake @@ -44,7 +44,7 @@ if(MSVC) set(BUILD_TARGET ${CMAKE_VS_PLATFORM_NAME}) else() execute_process( - COMMAND sh -c "$@ --version | head -1" _ ${CMAKE_C_COMPILER} + COMMAND sh -c "\"$@\" --version | head -1" _ ${CMAKE_C_COMPILER} OUTPUT_VARIABLE OUT OUTPUT_STRIP_TRAILING_WHITESPACE ) diff --git a/common/CMakeLists.txt b/common/CMakeLists.txt index df1cdf9a5..24b7f8741 100644 --- a/common/CMakeLists.txt +++ b/common/CMakeLists.txt @@ -56,6 +56,7 @@ add_library(${TARGET} STATIC arg.cpp arg.h base64.hpp + chat-template.hpp common.cpp common.h console.cpp @@ -64,6 +65,7 @@ add_library(${TARGET} STATIC json.hpp log.cpp log.h + minja.hpp ngram-cache.cpp ngram-cache.h sampling.cpp diff --git a/common/arg.cpp b/common/arg.cpp index c81b15217..a6226a34b 100644 --- a/common/arg.cpp +++ b/common/arg.cpp @@ -22,6 +22,11 @@ common_arg & common_arg::set_examples(std::initializer_list return *this; } +common_arg & common_arg::set_excludes(std::initializer_list excludes) { + this->excludes = std::move(excludes); + return *this; +} + common_arg & common_arg::set_env(const char * env) { help = help + "\n(env: " + env + ")"; this->env = env; @@ -37,6 +42,10 @@ bool common_arg::in_example(enum llama_example ex) { return examples.find(ex) != examples.end(); } +bool common_arg::is_exclude(enum llama_example ex) { + return excludes.find(ex) != excludes.end(); +} + bool common_arg::get_value_from_env(std::string & output) { if (env == nullptr) return false; char * value = std::getenv(env); @@ -121,17 +130,27 @@ std::string common_arg::to_string() { static void common_params_handle_model_default( std::string & model, - std::string & model_url, + const std::string & model_url, std::string & hf_repo, - std::string & hf_file) { + std::string & hf_file, + const std::string & hf_token, + const std::string & model_default) { if (!hf_repo.empty()) { // short-hand to avoid specifying --hf-file -> default it to --model if (hf_file.empty()) { if (model.empty()) { - throw std::invalid_argument("error: --hf-repo requires either --hf-file or --model\n"); + auto auto_detected = common_get_hf_file(hf_repo, hf_token); + if (auto_detected.first.empty() || auto_detected.second.empty()) { + exit(1); // built without CURL, error message already printed + } + hf_repo = auto_detected.first; + hf_file = auto_detected.second; + } else { + hf_file = model; } - hf_file = model; - } else if (model.empty()) { + } + // make sure model path is present (for caching purposes) + if (model.empty()) { // this is to avoid different repo having same file name, or same file name in different subdirs std::string filename = hf_repo + "_" + hf_file; // to make sure we don't have any slashes in the filename @@ -145,7 +164,7 @@ static void common_params_handle_model_default( model = fs_get_cache_file(string_split(f, '/').back()); } } else if (model.empty()) { - model = DEFAULT_MODEL_PATH; + model = model_default; } } @@ -281,8 +300,9 @@ static bool common_params_parse_ex(int argc, char ** argv, common_params_context } // TODO: refactor model params in a common struct - common_params_handle_model_default(params.model, params.model_url, params.hf_repo, params.hf_file); - common_params_handle_model_default(params.vocoder.model, params.vocoder.model_url, params.vocoder.hf_repo, params.vocoder.hf_file); + common_params_handle_model_default(params.model, params.model_url, params.hf_repo, params.hf_file, params.hf_token, DEFAULT_MODEL_PATH); + common_params_handle_model_default(params.speculative.model, params.speculative.model_url, params.speculative.hf_repo, params.speculative.hf_file, params.hf_token, ""); + common_params_handle_model_default(params.vocoder.model, params.vocoder.model_url, params.vocoder.hf_repo, params.vocoder.hf_file, params.hf_token, ""); if (params.escape) { string_process_escapes(params.prompt); @@ -305,6 +325,14 @@ static bool common_params_parse_ex(int argc, char ** argv, common_params_context throw std::invalid_argument("error: either --embedding or --reranking can be specified, but not both"); } + if (!params.chat_template.empty() && !common_chat_verify_template(params.chat_template, params.use_jinja)) { + throw std::runtime_error(string_format( + "error: the supplied chat template is not supported: %s%s\n", + params.chat_template.c_str(), + params.use_jinja ? "" : "\nnote: llama.cpp was started without --jinja, we only support commonly used templates" + )); + } + return true; } @@ -358,6 +386,30 @@ static std::vector parse_device_list(const std::string & val return devices; } +static void add_rpc_devices(std::string servers) { + auto rpc_servers = string_split(servers, ','); + if (rpc_servers.empty()) { + throw std::invalid_argument("no RPC servers specified"); + } + ggml_backend_reg_t rpc_reg = ggml_backend_reg_by_name("RPC"); + if (!rpc_reg) { + throw std::invalid_argument("failed to find RPC backend"); + } + typedef ggml_backend_dev_t (*ggml_backend_rpc_add_device_t)(const char * endpoint); + ggml_backend_rpc_add_device_t ggml_backend_rpc_add_device_fn = (ggml_backend_rpc_add_device_t) ggml_backend_reg_get_proc_address(rpc_reg, "ggml_backend_rpc_add_device"); + if (!ggml_backend_rpc_add_device_fn) { + throw std::invalid_argument("failed to find RPC device add function"); + } + for (const auto & server : rpc_servers) { + ggml_backend_dev_t dev = ggml_backend_rpc_add_device_fn(server.c_str()); + if (dev) { + ggml_backend_device_register(dev); + } else { + throw std::invalid_argument("failed to register RPC device"); + } + } +} + bool common_params_parse(int argc, char ** argv, common_params & params, llama_example ex, void(*print_usage)(int, char **)) { auto ctx_arg = common_params_parser_init(params, ex, print_usage); const common_params params_org = ctx_arg.params; // the example can modify the default params @@ -420,7 +472,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex * - if both {LLAMA_EXAMPLE_COMMON, LLAMA_EXAMPLE_*,} are set, we will prioritize the LLAMA_EXAMPLE_* matching current example */ auto add_opt = [&](common_arg arg) { - if (arg.in_example(ex) || arg.in_example(LLAMA_EXAMPLE_COMMON)) { + if ((arg.in_example(ex) || arg.in_example(LLAMA_EXAMPLE_COMMON)) && !arg.is_exclude(ex)) { ctx_arg.options.push_back(std::move(arg)); } }; @@ -649,7 +701,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex [](common_params & params, const std::string & value) { params.prompt = value; } - )); + ).set_excludes({LLAMA_EXAMPLE_SERVER})); add_opt(common_arg( {"--no-perf"}, string_format("disable internal libllama performance timings (default: %s)", params.no_perf ? "true" : "false"), @@ -673,7 +725,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex params.prompt.pop_back(); } } - )); + ).set_excludes({LLAMA_EXAMPLE_SERVER})); add_opt(common_arg( {"--in-file"}, "FNAME", "an input file (repeat to specify multiple files)", @@ -700,7 +752,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex params.prompt = ss.str(); fprintf(stderr, "Read %zu bytes from binary file %s\n", params.prompt.size(), value.c_str()); } - )); + ).set_excludes({LLAMA_EXAMPLE_SERVER})); add_opt(common_arg( {"-e", "--escape"}, string_format("process escapes sequences (\\n, \\r, \\t, \\', \\\", \\\\) (default: %s)", params.escape ? "true" : "false"), @@ -759,15 +811,19 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_SERVER})); add_opt(common_arg( {"-cnv", "--conversation"}, - string_format( - "run in conversation mode:\n" - "- does not print special tokens and suffix/prefix\n" - "- interactive mode is also enabled\n" - "(default: %s)", - params.conversation ? "true" : "false" - ), + "run in conversation mode:\n" + "- does not print special tokens and suffix/prefix\n" + "- interactive mode is also enabled\n" + "(default: auto enabled if chat template is available)", [](common_params & params) { - params.conversation = true; + params.conversation_mode = COMMON_CONVERSATION_MODE_ENABLED; + } + ).set_examples({LLAMA_EXAMPLE_MAIN})); + add_opt(common_arg( + {"-no-cnv", "--no-conversation"}, + "force disable conversation mode (default: false)", + [](common_params & params) { + params.conversation_mode = COMMON_CONVERSATION_MODE_DISABLED; } ).set_examples({LLAMA_EXAMPLE_MAIN})); add_opt(common_arg( @@ -1363,7 +1419,8 @@ common_params_context common_params_parser_init(common_params & params, llama_ex {"--rpc"}, "SERVERS", "comma separated list of RPC servers", [](common_params & params, const std::string & value) { - params.rpc_servers = value; + add_rpc_devices(value); + GGML_UNUSED(params); } ).set_env("LLAMA_ARG_RPC")); } @@ -1574,21 +1631,30 @@ common_params_context common_params_parser_init(common_params & params, llama_ex } ).set_env("LLAMA_ARG_MODEL_URL")); add_opt(common_arg( - {"-hfr", "--hf-repo"}, "REPO", - "Hugging Face model repository (default: unused)", + {"-hf", "-hfr", "--hf-repo"}, "/[:quant]", + "Hugging Face model repository; quant is optional, case-insensitive, default to Q4_K_M, or falls back to the first file in the repo if Q4_K_M doesn't exist.\n" + "example: unsloth/phi-4-GGUF:q4_k_m\n" + "(default: unused)", [](common_params & params, const std::string & value) { params.hf_repo = value; } ).set_env("LLAMA_ARG_HF_REPO")); + add_opt(common_arg( + {"-hfd", "-hfrd", "--hf-repo-draft"}, "/[:quant]", + "Same as --hf-repo, but for the draft model (default: unused)", + [](common_params & params, const std::string & value) { + params.speculative.hf_repo = value; + } + ).set_env("LLAMA_ARG_HFD_REPO")); add_opt(common_arg( {"-hff", "--hf-file"}, "FILE", - "Hugging Face model file (default: unused)", + "Hugging Face model file. If specified, it will override the quant in --hf-repo (default: unused)", [](common_params & params, const std::string & value) { params.hf_file = value; } ).set_env("LLAMA_ARG_HF_FILE")); add_opt(common_arg( - {"-hfrv", "--hf-repo-v"}, "REPO", + {"-hfv", "-hfrv", "--hf-repo-v"}, "/[:quant]", "Hugging Face model repository for the vocoder model (default: unused)", [](common_params & params, const std::string & value) { params.vocoder.hf_repo = value; @@ -1889,24 +1955,44 @@ common_params_context common_params_parser_init(common_params & params, llama_ex } } ).set_examples({LLAMA_EXAMPLE_SERVER})); + add_opt(common_arg( + {"--jinja"}, + "use jinja template for chat (default: disabled)", + [](common_params & params) { + params.use_jinja = true; + } + ).set_examples({LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_MAIN}).set_env("LLAMA_ARG_JINJA")); add_opt(common_arg( {"--chat-template"}, "JINJA_TEMPLATE", string_format( "set custom jinja chat template (default: template taken from model's metadata)\n" "if suffix/prefix are specified, template will be disabled\n" + "only commonly used templates are accepted (unless --jinja is set before this flag):\n" "list of built-in templates:\n%s", list_builtin_chat_templates().c_str() ), [](common_params & params, const std::string & value) { - if (!common_chat_verify_template(value)) { - throw std::runtime_error(string_format( - "error: the supplied chat template is not supported: %s\n" - "note: llama.cpp does not use jinja parser, we only support commonly used templates\n", - value.c_str() - )); - } params.chat_template = value; } ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_CHAT_TEMPLATE")); + add_opt(common_arg( + {"--chat-template-file"}, "JINJA_TEMPLATE_FILE", + string_format( + "set custom jinja chat template file (default: template taken from model's metadata)\n" + "if suffix/prefix are specified, template will be disabled\n" + "only commonly used templates are accepted (unless --jinja is set before this flag):\n" + "list of built-in templates:\n%s", list_builtin_chat_templates().c_str() + ), + [](common_params & params, const std::string & value) { + std::ifstream file(value); + if (!file) { + throw std::runtime_error(string_format("error: failed to open file '%s'\n", value.c_str())); + } + std::copy( + std::istreambuf_iterator(file), + std::istreambuf_iterator(), + std::back_inserter(params.chat_template)); + } + ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_CHAT_TEMPLATE_FILE")); add_opt(common_arg( {"-sps", "--slot-prompt-similarity"}, "SIMILARITY", string_format("how much the prompt of a request must match the prompt of a slot in order to use that slot (default: %.2f, 0.0 = disabled)\n", params.slot_prompt_similarity), @@ -2205,6 +2291,13 @@ common_params_context common_params_parser_init(common_params & params, llama_ex params.vocoder.model = value; } ).set_examples({LLAMA_EXAMPLE_TTS, LLAMA_EXAMPLE_SERVER})); + add_opt(common_arg( + {"--tts-use-guide-tokens"}, + "Use guide tokens to improve TTS word recall", + [](common_params & params) { + params.vocoder.use_guide_tokens = true; + } + ).set_examples({LLAMA_EXAMPLE_TTS, LLAMA_EXAMPLE_SERVER})); // model-specific add_opt(common_arg( diff --git a/common/arg.h b/common/arg.h index a6700d323..49ab8667b 100644 --- a/common/arg.h +++ b/common/arg.h @@ -12,6 +12,7 @@ struct common_arg { std::set examples = {LLAMA_EXAMPLE_COMMON}; + std::set excludes = {}; std::vector args; const char * value_hint = nullptr; // help text or example for arg value const char * value_hint_2 = nullptr; // for second arg value @@ -53,9 +54,11 @@ struct common_arg { ) : args(args), value_hint(value_hint), value_hint_2(value_hint_2), help(help), handler_str_str(handler) {} common_arg & set_examples(std::initializer_list examples); + common_arg & set_excludes(std::initializer_list excludes); common_arg & set_env(const char * env); common_arg & set_sparam(); bool in_example(enum llama_example ex); + bool is_exclude(enum llama_example ex); bool get_value_from_env(std::string & output); bool has_value_from_env(); std::string to_string(); diff --git a/common/chat-template.hpp b/common/chat-template.hpp new file mode 100644 index 000000000..42ee0b615 --- /dev/null +++ b/common/chat-template.hpp @@ -0,0 +1,268 @@ +/* + Copyright 2024 Google LLC + + Use of this source code is governed by an MIT-style + license that can be found in the LICENSE file or at + https://opensource.org/licenses/MIT. +*/ +// SPDX-License-Identifier: MIT +#pragma once + +#include "minja.hpp" +#include +#include +#include + +using json = nlohmann::ordered_json; + +namespace minja { + +class chat_template { + public: + + private: + bool supports_tools_ = true; + // Meta-Llama-3.1-8B-Instruct's template expects arguments to be an object. + // Most other templates (and OpenAI's API) expect the arguments object to be stringified. + bool requires_object_arguments_ = false; + bool requires_typed_content_ = false; + bool supports_system_role_ = true; + bool supports_parallel_tool_calls_ = false; + std::string source_; + std::string bos_token_; + std::string eos_token_; + std::shared_ptr template_root_; + + std::string try_raw_render( + const nlohmann::ordered_json & messages, + const nlohmann::ordered_json & tools, + bool add_generation_prompt, + const nlohmann::ordered_json & extra_context = nlohmann::ordered_json()) const + { + try { + auto prompt = apply(messages, tools, add_generation_prompt, extra_context, /* adjust_inputs= */ false); + // fprintf(stderr, "Prompt: %s\n", prompt.c_str()); + return prompt; + } catch (const std::exception & e) { + // fprintf(stderr, "Error: %s\n", e.what()); + return ""; + } + } + + public: + chat_template(const std::string & source, const std::string & bos_token, const std::string & eos_token) + : source_(source), bos_token_(bos_token), eos_token_(eos_token) + { + template_root_ = minja::Parser::parse(source_, { + /* .trim_blocks = */ true, + /* .lstrip_blocks = */ true, + /* .keep_trailing_newline = */ false, + }); + supports_tools_ = source.find("tools") != std::string::npos; + + auto renders_string_arguments = + try_raw_render({ + { + {"role", "user"}, + {"content", "Hey"} + }, + { + {"role", "assistant"}, + {"tool_calls", json::array({ + { + {"id", "call_1___"}, + {"type", "function"}, + {"function", { + {"arguments", "{\"code\": \"print('Hello, World!')\"}"}, + {"name", "ipython"}, + }}, + }, + })}, + } + }, {}, false).find("{\"code\": \"print") != std::string::npos; + if (!renders_string_arguments) { + auto renders_object_arguments = + try_raw_render({ + { + {"role", "user"}, + {"content", "Hey"} + }, + { + {"role", "assistant"}, + {"tool_calls", json::array({ + { + {"id", "call_1___"}, + {"type", "function"}, + {"function", { + {"arguments", { + {"code", "print('Hello, World!')"}, + }}, + {"name", "ipython"}, + }}, + }, + })}, + } + }, {}, false).find("{\"code\": \"print") != std::string::npos; + requires_object_arguments_ = renders_object_arguments; + } + supports_parallel_tool_calls_ = source.find("tool_call_id") != std::string::npos; + + supports_system_role_ = try_raw_render({ + {{"role", "system"}, {"content", ""}}, + {{"role", "user"}, {"content", "Hey"}} + }, {}, false).find("") != std::string::npos; + + requires_typed_content_ = try_raw_render({{{"role", "user"}, {"content", "Hey"}}}, {}, false).find("Hey") == std::string::npos + && try_raw_render({{{"role", "user"}, {"content", {{{"type", "text"}, {"text", "Hey"}}}}}}, {}, false).find("Hey") != std::string::npos; + } + + const std::string & source() const { return source_; } + const std::string & bos_token() const { return bos_token_; } + const std::string & eos_token() const { return eos_token_; } + bool supports_tools() const { return supports_tools_; } + bool supports_parallel_tool_calls() const { return supports_parallel_tool_calls_; } + + std::string apply( + const nlohmann::ordered_json & messages, + const nlohmann::ordered_json & tools, + bool add_generation_prompt, + const nlohmann::ordered_json & extra_context = nlohmann::ordered_json(), + bool adjust_inputs = true) const + { + json actual_messages; + + // First, "fix" messages so they have a chance to be rendered correctly by the template + + if (adjust_inputs && (requires_object_arguments_ || !supports_system_role_ || !supports_tools_ || requires_typed_content_)) { + actual_messages = json::array(); + + auto add_message = [&](const json & msg) { + if (requires_typed_content_ && msg.contains("content") && !msg.at("content").is_null() && msg.at("content").is_string()) { + actual_messages.push_back({ + {"role", msg.at("role")}, + {"content", {{ + {"type", "text"}, + {"text", msg.at("content")}, + }}}, + }); + } else { + actual_messages.push_back(msg); + } + }; + + std::string pending_system; + auto flush_sys = [&]() { + if (!pending_system.empty()) { + add_message({ + {"role", "user"}, + {"content", pending_system}, + }); + pending_system.clear(); + } + }; + for (const auto & message_ : messages) { + auto message = message_; + if (!message.contains("role") || !message.contains("content")) { + throw std::runtime_error("message must have 'role' and 'content' fields: " + message.dump()); + } + std::string role = message.at("role"); + + if (message.contains("tool_calls")) { + if (requires_object_arguments_ || !supports_tools_) { + for (auto & tool_call : message.at("tool_calls")) { + if (tool_call["type"] == "function") { + auto & function = tool_call.at("function"); + std::string arguments = function.at("arguments"); + function["arguments"] = json::parse(arguments); + } + } + } + if (!supports_tools_) { + auto content = message.at("content"); + auto tool_calls = json::array(); + for (const auto & tool_call : message.at("tool_calls")) { + if (tool_call.at("type") != "function") { + continue; + } + const auto & function = tool_call.at("function"); + auto tc = json { + {"name", function.at("name")}, + {"arguments", function.at("arguments")}, + }; + if (tool_call.contains("id")) { + tc["id"] = tool_call["id"]; + } + tool_calls.push_back(tc); + } + auto obj = json { + {"tool_calls", tool_calls}, + }; + if (!content.is_null() && content != "") { + obj["content"] = content; + } + message["content"] = obj.dump(2); + message.erase("tool_calls"); + } + } + if (!supports_tools_ && role == "tool") { + message["role"] = "user"; + auto obj = json { + {"tool_response", { + {"tool", message.at("name")}, + {"content", message.at("content")}, + }}, + }; + if (message.contains("tool_call_id")) { + obj["tool_response"]["tool_call_id"] = message.at("tool_call_id"); + } + message["content"] = obj.dump(2); + message.erase("name"); + } + + if (!message["content"].is_null() && !supports_system_role_) { + std::string content = message.at("content"); + if (role == "system") { + if (!pending_system.empty()) pending_system += "\n"; + pending_system += content; + continue; + } else { + if (role == "user") { + if (!pending_system.empty()) { + message["content"] = pending_system + (content.empty() ? "" : "\n" + content); + pending_system.clear(); + } + } else { + flush_sys(); + } + } + } + add_message(message); + } + flush_sys(); + } else { + actual_messages = messages; + } + + auto context = minja::Context::make(json({ + {"messages", actual_messages}, + {"add_generation_prompt", add_generation_prompt}, + {"bos_token", bos_token_}, + {"eos_token", eos_token_}, + })); + + if (!tools.is_null()) { + auto tools_val = minja::Value(tools); + context->set("tools", tools_val); + } + if (!extra_context.is_null()) { + for (auto & kv : extra_context.items()) { + minja::Value val(kv.value()); + context->set(kv.key(), val); + } + } + + return template_root_->render(context); + } +}; + +} // namespace minja diff --git a/common/common.cpp b/common/common.cpp index 4bb140ee2..6dea8e3d2 100644 --- a/common/common.cpp +++ b/common/common.cpp @@ -2,6 +2,9 @@ #define _SILENCE_CXX17_CODECVT_HEADER_DEPRECATION_WARNING #endif +#include "ggml.h" +#include "gguf.h" + #include "common.h" #include "log.h" // Change JSON_ASSERT from assert() to GGML_ASSERT: @@ -9,6 +12,7 @@ #include "json.hpp" #include "json-schema-to-grammar.h" #include "llama.h" +#include "chat-template.hpp" #include #include @@ -70,6 +74,22 @@ #include #endif #define LLAMA_CURL_MAX_URL_LENGTH 2084 // Maximum URL Length in Chrome: 2083 + +// +// CURL utils +// + +using curl_ptr = std::unique_ptr; + +// cannot use unique_ptr for curl_slist, because we cannot update without destroying the old one +struct curl_slist_ptr { + struct curl_slist * ptr = nullptr; + ~curl_slist_ptr() { + if (ptr) { + curl_slist_free_all(ptr); + } + } +}; #endif // LLAMA_USE_CURL using json = nlohmann::ordered_json; @@ -464,6 +484,48 @@ void string_replace_all(std::string & s, const std::string & search, const std:: s = std::move(builder); } +std::string string_join(const std::vector & values, const std::string & separator) { + std::ostringstream result; + for (size_t i = 0; i < values.size(); ++i) { + if (i > 0) { + result << separator; + } + result << values[i]; + } + return result.str(); +} + +std::vector string_split(const std::string & str, const std::string & delimiter) { + std::vector parts; + size_t start = 0; + size_t end = str.find(delimiter); + + while (end != std::string::npos) { + parts.push_back(str.substr(start, end - start)); + start = end + delimiter.length(); + end = str.find(delimiter, start); + } + + parts.push_back(str.substr(start)); + + return parts; +} + +std::string string_repeat(const std::string & str, size_t n) { + if (n == 0) { + return ""; + } + + std::string result; + result.reserve(str.length() * n); + + for (size_t i = 0; i < n; ++i) { + result += str; + } + + return result; +} + std::string string_from(bool value) { return value ? "true" : "false"; } @@ -846,7 +908,7 @@ struct common_init_result common_init_from_params(common_params & params) { } else if (!params.model_url.empty()) { model = common_load_model_from_url(params.model_url, params.model, params.hf_token, mparams); } else { - model = llama_load_model_from_file(params.model.c_str(), mparams); + model = llama_model_load_from_file(params.model.c_str(), mparams); } if (model == NULL) { @@ -854,26 +916,28 @@ struct common_init_result common_init_from_params(common_params & params) { return iparams; } + const llama_vocab * vocab = llama_model_get_vocab(model); + if (params.reranking) { bool ok = true; - if (llama_token_bos(model) == LLAMA_TOKEN_NULL) { - LOG_WRN("%s: warning: model does not have a BOS token, reranking will not work\n", __func__); + if (llama_vocab_bos(vocab) == LLAMA_TOKEN_NULL) { + LOG_WRN("%s: warning: vocab does not have a BOS token, reranking will not work\n", __func__); ok = false; } - if (llama_token_eos(model) == LLAMA_TOKEN_NULL) { - LOG_WRN("%s: warning: model does not have an EOS token, reranking will not work\n", __func__); + if (llama_vocab_eos(vocab) == LLAMA_TOKEN_NULL) { + LOG_WRN("%s: warning: vocab does not have an EOS token, reranking will not work\n", __func__); ok = false; } - if (llama_token_sep(model) == LLAMA_TOKEN_NULL) { - LOG_WRN("%s: warning: model does not have a SEP token, reranking will not work\n", __func__); + if (llama_vocab_sep(vocab) == LLAMA_TOKEN_NULL) { + LOG_WRN("%s: warning: vocab does not have a SEP token, reranking will not work\n", __func__); ok = false; } if (!ok) { - llama_free_model(model); + llama_model_free(model); return iparams; } @@ -881,10 +945,10 @@ struct common_init_result common_init_from_params(common_params & params) { auto cparams = common_context_params_to_llama(params); - llama_context * lctx = llama_new_context_with_model(model, cparams); + llama_context * lctx = llama_init_from_model(model, cparams); if (lctx == NULL) { LOG_ERR("%s: failed to create context with model '%s'\n", __func__, params.model.c_str()); - llama_free_model(model); + llama_model_free(model); return iparams; } @@ -895,25 +959,26 @@ struct common_init_result common_init_from_params(common_params & params) { if (!params.control_vectors.empty()) { if (params.control_vector_layer_start <= 0) params.control_vector_layer_start = 1; - if (params.control_vector_layer_end <= 0) params.control_vector_layer_end = llama_n_layer(model); + if (params.control_vector_layer_end <= 0) params.control_vector_layer_end = llama_model_n_layer(model); const auto cvec = common_control_vector_load(params.control_vectors); if (cvec.n_embd == -1) { llama_free(lctx); - llama_free_model(model); + llama_model_free(model); return iparams; } - int err = llama_control_vector_apply(lctx, - cvec.data.data(), - cvec.data.size(), - cvec.n_embd, - params.control_vector_layer_start, - params.control_vector_layer_end); + int err = llama_apply_adapter_cvec( + lctx, + cvec.data.data(), + cvec.data.size(), + cvec.n_embd, + params.control_vector_layer_start, + params.control_vector_layer_end); if (err) { llama_free(lctx); - llama_free_model(model); + llama_model_free(model); return iparams; } @@ -921,12 +986,12 @@ struct common_init_result common_init_from_params(common_params & params) { // load and optionally apply lora adapters for (auto & la : params.lora_adapters) { - llama_lora_adapter_ptr lora; - lora.reset(llama_lora_adapter_init(model, la.path.c_str())); + llama_adapter_lora_ptr lora; + lora.reset(llama_adapter_lora_init(model, la.path.c_str())); if (lora == nullptr) { LOG_ERR("%s: failed to apply lora adapter '%s'\n", __func__, la.path.c_str()); llama_free(lctx); - llama_free_model(model); + llama_model_free(model); return iparams; } @@ -935,17 +1000,17 @@ struct common_init_result common_init_from_params(common_params & params) { } if (!params.lora_init_without_apply) { - common_lora_adapters_apply(lctx, params.lora_adapters); + common_set_adapter_lora(lctx, params.lora_adapters); } - if (params.sampling.ignore_eos && llama_token_eos(model) == LLAMA_TOKEN_NULL) { - LOG_WRN("%s: warning: model does not have an EOS token, ignoring --ignore-eos\n", __func__); + if (params.sampling.ignore_eos && llama_vocab_eos(vocab) == LLAMA_TOKEN_NULL) { + LOG_WRN("%s: warning: vocab does not have an EOS token, ignoring --ignore-eos\n", __func__); params.sampling.ignore_eos = false; } if (params.sampling.ignore_eos) { - for (llama_token i = 0; i < llama_n_vocab(model); i++) { - if (llama_token_is_eog(model, i)) { + for (llama_token i = 0; i < llama_vocab_n_tokens(vocab); i++) { + if (llama_vocab_is_eog(vocab, i)) { LOG_INF("%s: added %s logit bias = %f\n", __func__, common_token_to_piece(lctx, i).c_str(), -INFINITY); params.sampling.logit_bias.push_back({i, -INFINITY}); } @@ -966,8 +1031,9 @@ struct common_init_result common_init_from_params(common_params & params) { LOG_WRN("%s: warming up the model with an empty run - please wait ... (--no-warmup to disable)\n", __func__); std::vector tmp; - llama_token bos = llama_token_bos(model); - llama_token eos = llama_token_eos(model); + llama_token bos = llama_vocab_bos(vocab); + llama_token eos = llama_vocab_eos(vocab); + // some models (e.g. T5) don't have a BOS token if (bos != LLAMA_TOKEN_NULL) { tmp.push_back(bos); @@ -982,7 +1048,7 @@ struct common_init_result common_init_from_params(common_params & params) { if (llama_model_has_encoder(model)) { llama_encode(lctx, llama_batch_get_one(tmp.data(), tmp.size())); llama_token decoder_start_token_id = llama_model_decoder_start_token(model); - if (decoder_start_token_id == -1) { + if (decoder_start_token_id == LLAMA_TOKEN_NULL) { decoder_start_token_id = bos; } tmp.clear(); @@ -1002,11 +1068,11 @@ struct common_init_result common_init_from_params(common_params & params) { return iparams; } -void common_lora_adapters_apply(struct llama_context * ctx, std::vector & lora) { - llama_lora_adapter_clear(ctx); +void common_set_adapter_lora(struct llama_context * ctx, std::vector & lora) { + llama_clear_adapter_lora(ctx); for (auto & la : lora) { if (la.scale != 0.0f) { - llama_lora_adapter_set(ctx, la.ptr, la.scale); + llama_set_adapter_lora(ctx, la.ptr, la.scale); } } } @@ -1020,7 +1086,6 @@ struct llama_model_params common_model_params_to_llama(common_params & params) { if (params.n_gpu_layers != -1) { mparams.n_gpu_layers = params.n_gpu_layers; } - mparams.rpc_servers = params.rpc_servers.c_str(); mparams.main_gpu = params.main_gpu; mparams.split_mode = params.split_mode; mparams.tensor_split = params.tensor_split; @@ -1123,7 +1188,8 @@ static bool curl_perform_with_retry(const std::string & url, CURL * curl, int ma static bool common_download_file(const std::string & url, const std::string & path, const std::string & hf_token) { // Initialize libcurl - std::unique_ptr curl(curl_easy_init(), &curl_easy_cleanup); + curl_ptr curl(curl_easy_init(), &curl_easy_cleanup); + curl_slist_ptr http_headers; if (!curl) { LOG_ERR("%s: error initializing libcurl\n", __func__); return false; @@ -1137,11 +1203,9 @@ static bool common_download_file(const std::string & url, const std::string & pa // Check if hf-token or bearer-token was specified if (!hf_token.empty()) { - std::string auth_header = "Authorization: Bearer "; - auth_header += hf_token.c_str(); - struct curl_slist *http_headers = NULL; - http_headers = curl_slist_append(http_headers, auth_header.c_str()); - curl_easy_setopt(curl.get(), CURLOPT_HTTPHEADER, http_headers); + std::string auth_header = "Authorization: Bearer " + hf_token; + http_headers.ptr = curl_slist_append(http_headers.ptr, auth_header.c_str()); + curl_easy_setopt(curl.get(), CURLOPT_HTTPHEADER, http_headers.ptr); } #if defined(_WIN32) @@ -1411,7 +1475,7 @@ struct llama_model * common_load_model_from_url( } } - return llama_load_model_from_file(local_path.c_str(), params); + return llama_model_load_from_file(local_path.c_str(), params); } struct llama_model * common_load_model_from_hf( @@ -1437,6 +1501,80 @@ struct llama_model * common_load_model_from_hf( return common_load_model_from_url(model_url, local_path, hf_token, params); } +/** + * Allow getting the HF file from the HF repo with tag (like ollama), for example: + * - bartowski/Llama-3.2-3B-Instruct-GGUF:q4 + * - bartowski/Llama-3.2-3B-Instruct-GGUF:Q4_K_M + * - bartowski/Llama-3.2-3B-Instruct-GGUF:q5_k_s + * Tag is optional, default to "latest" (meaning it checks for Q4_K_M first, then Q4, then if not found, return the first GGUF file in repo) + * + * Return pair of (with "repo" already having tag removed) + * + * Note: we use the Ollama-compatible HF API, but not using the blobId. Instead, we use the special "ggufFile" field which returns the value for "hf_file". This is done to be backward-compatible with existing cache files. + */ +std::pair common_get_hf_file(const std::string & hf_repo_with_tag, const std::string & hf_token) { + auto parts = string_split(hf_repo_with_tag, ':'); + std::string tag = parts.size() > 1 ? parts.back() : "latest"; + std::string hf_repo = parts[0]; + if (string_split(hf_repo, '/').size() != 2) { + throw std::invalid_argument("error: invalid HF repo format, expected /[:quant]\n"); + } + + // fetch model info from Hugging Face Hub API + json model_info; + curl_ptr curl(curl_easy_init(), &curl_easy_cleanup); + curl_slist_ptr http_headers; + std::string res_str; + std::string url = "https://huggingface.co/v2/" + hf_repo + "/manifests/" + tag; + curl_easy_setopt(curl.get(), CURLOPT_URL, url.c_str()); + curl_easy_setopt(curl.get(), CURLOPT_NOPROGRESS, 1L); + typedef size_t(*CURLOPT_WRITEFUNCTION_PTR)(void * ptr, size_t size, size_t nmemb, void * data); + auto write_callback = [](void * ptr, size_t size, size_t nmemb, void * data) -> size_t { + static_cast(data)->append((char * ) ptr, size * nmemb); + return size * nmemb; + }; + curl_easy_setopt(curl.get(), CURLOPT_WRITEFUNCTION, static_cast(write_callback)); + curl_easy_setopt(curl.get(), CURLOPT_WRITEDATA, &res_str); +#if defined(_WIN32) + curl_easy_setopt(curl.get(), CURLOPT_SSL_OPTIONS, CURLSSLOPT_NATIVE_CA); +#endif + if (!hf_token.empty()) { + std::string auth_header = "Authorization: Bearer " + hf_token; + http_headers.ptr = curl_slist_append(http_headers.ptr, auth_header.c_str()); + } + // Important: the User-Agent must be "llama-cpp" to get the "ggufFile" field in the response + http_headers.ptr = curl_slist_append(http_headers.ptr, "User-Agent: llama-cpp"); + http_headers.ptr = curl_slist_append(http_headers.ptr, "Accept: application/json"); + curl_easy_setopt(curl.get(), CURLOPT_HTTPHEADER, http_headers.ptr); + + CURLcode res = curl_easy_perform(curl.get()); + + if (res != CURLE_OK) { + throw std::runtime_error("error: cannot make GET request to HF API"); + } + + long res_code; + curl_easy_getinfo(curl.get(), CURLINFO_RESPONSE_CODE, &res_code); + if (res_code == 200) { + model_info = json::parse(res_str); + } else if (res_code == 401) { + throw std::runtime_error("error: model is private or does not exist; if you are accessing a gated model, please provide a valid HF token"); + } else { + throw std::runtime_error(string_format("error from HF API, response code: %ld, data: %s", res_code, res_str.c_str())); + } + + // check response + if (!model_info.contains("ggufFile")) { + throw std::runtime_error("error: model does not have ggufFile"); + } + json & gguf_file = model_info.at("ggufFile"); + if (!gguf_file.contains("rfilename")) { + throw std::runtime_error("error: ggufFile does not have rfilename"); + } + + return std::make_pair(hf_repo, gguf_file.at("rfilename")); +} + #else struct llama_model * common_load_model_from_url( @@ -1458,6 +1596,11 @@ struct llama_model * common_load_model_from_hf( return nullptr; } +std::pair common_get_hf_file(const std::string &, const std::string &) { + LOG_WRN("%s: llama.cpp built without libcurl, downloading from Hugging Face not supported.\n", __func__); + return std::make_pair("", ""); +} + #endif // LLAMA_USE_CURL // @@ -1556,21 +1699,23 @@ std::vector common_tokenize( const std::string & text, bool add_special, bool parse_special) { - return common_tokenize(llama_get_model(ctx), text, add_special, parse_special); + const llama_model * model = llama_get_model(ctx); + const llama_vocab * vocab = llama_model_get_vocab(model); + return common_tokenize(vocab, text, add_special, parse_special); } std::vector common_tokenize( - const struct llama_model * model, + const struct llama_vocab * vocab, const std::string & text, bool add_special, bool parse_special) { // upper limit for the number of tokens int n_tokens = text.length() + 2 * add_special; std::vector result(n_tokens); - n_tokens = llama_tokenize(model, text.data(), text.length(), result.data(), result.size(), add_special, parse_special); + n_tokens = llama_tokenize(vocab, text.data(), text.length(), result.data(), result.size(), add_special, parse_special); if (n_tokens < 0) { result.resize(-n_tokens); - int check = llama_tokenize(model, text.data(), text.length(), result.data(), result.size(), add_special, parse_special); + int check = llama_tokenize(vocab, text.data(), text.length(), result.data(), result.size(), add_special, parse_special); GGML_ASSERT(check == -n_tokens); } else { result.resize(n_tokens); @@ -1579,12 +1724,18 @@ std::vector common_tokenize( } std::string common_token_to_piece(const struct llama_context * ctx, llama_token token, bool special) { + const llama_model * model = llama_get_model(ctx); + const llama_vocab * vocab = llama_model_get_vocab(model); + return common_token_to_piece(vocab, token, special); +} + +std::string common_token_to_piece(const struct llama_vocab * vocab, llama_token token, bool special) { std::string piece; piece.resize(piece.capacity()); // using string internal cache, 15 bytes + '\n' - const int n_chars = llama_token_to_piece(llama_get_model(ctx), token, &piece[0], piece.size(), 0, special); + const int n_chars = llama_token_to_piece(vocab, token, &piece[0], piece.size(), 0, special); if (n_chars < 0) { piece.resize(-n_chars); - int check = llama_token_to_piece(llama_get_model(ctx), token, &piece[0], piece.size(), 0, special); + int check = llama_token_to_piece(vocab, token, &piece[0], piece.size(), 0, special); GGML_ASSERT(check == -n_chars); } else { @@ -1594,13 +1745,19 @@ std::string common_token_to_piece(const struct llama_context * ctx, llama_token return piece; } -std::string common_detokenize(llama_context * ctx, const std::vector & tokens, bool special) { +std::string common_detokenize(const struct llama_context * ctx, const std::vector & tokens, bool special) { + const llama_model * model = llama_get_model(ctx); + const llama_vocab * vocab = llama_model_get_vocab(model); + return common_detokenize(vocab, tokens, special); +} + +std::string common_detokenize(const struct llama_vocab * vocab, const std::vector & tokens, bool special) { std::string text; text.resize(std::max(text.capacity(), tokens.size())); - int32_t n_chars = llama_detokenize(llama_get_model(ctx), tokens.data(), (int32_t)tokens.size(), &text[0], (int32_t)text.size(), false, special); + int32_t n_chars = llama_detokenize(vocab, tokens.data(), (int32_t)tokens.size(), &text[0], (int32_t)text.size(), false, special); if (n_chars < 0) { text.resize(-n_chars); - n_chars = llama_detokenize(llama_get_model(ctx), tokens.data(), (int32_t)tokens.size(), &text[0], (int32_t)text.size(), false, special); + n_chars = llama_detokenize(vocab, tokens.data(), (int32_t)tokens.size(), &text[0], (int32_t)text.size(), false, special); GGML_ASSERT(n_chars <= (int32_t)text.size()); // whitespace trimming is performed after per-token detokenization } @@ -1614,75 +1771,75 @@ std::string common_detokenize(llama_context * ctx, const std::vector 0) { - std::vector model_template(res + 1, 0); - llama_model_meta_val_str(model, template_key, model_template.data(), model_template.size()); - return std::string(model_template.data(), model_template.size() - 1); +bool common_chat_verify_template(const std::string & tmpl, bool use_jinja) { + if (use_jinja) { + try { + auto chat_template = minja::chat_template(tmpl, "", ""); + chat_template.apply({{ + {"role", "user"}, + {"content", "test"}, + }}, json(), true); + return true; + } catch (const std::exception & e) { + LOG_ERR("%s: failed to apply template: %s\n", __func__, e.what()); + return false; + } } - return ""; -} - -bool common_chat_verify_template(const std::string & tmpl) { llama_chat_message chat[] = {{"user", "test"}}; - int res = llama_chat_apply_template(nullptr, tmpl.c_str(), chat, 1, true, nullptr, 0); + const int res = llama_chat_apply_template(tmpl.c_str(), chat, 1, true, nullptr, 0); return res >= 0; } -std::string common_chat_apply_template(const struct llama_model * model, - const std::string & tmpl, +std::string common_chat_apply_template( + const common_chat_template & tmpl, const std::vector & msgs, - bool add_ass) { + bool add_ass, + bool use_jinja) { + if (use_jinja) { + auto messages = json::array(); + for (const auto & msg : msgs) { + messages.push_back({{"role", msg.role}, {"content", msg.content}}); + } + return tmpl.apply(messages, /* tools= */ json(), add_ass); + } + int alloc_size = 0; - bool fallback = false; // indicate if we must fallback to default chatml std::vector chat; - for (auto & msg : msgs) { + for (const auto & msg : msgs) { chat.push_back({msg.role.c_str(), msg.content.c_str()}); alloc_size += (msg.role.size() + msg.content.size()) * 1.25; } - const char * ptr_tmpl = tmpl.empty() ? nullptr : tmpl.c_str(); std::vector buf(alloc_size); // run the first time to get the total output length - int32_t res = llama_chat_apply_template(model, ptr_tmpl, chat.data(), chat.size(), add_ass, buf.data(), buf.size()); + int32_t res = llama_chat_apply_template(tmpl.source().c_str(), chat.data(), chat.size(), add_ass, buf.data(), buf.size()); // error: chat template is not supported if (res < 0) { - if (ptr_tmpl != nullptr) { - // if the custom "tmpl" is not supported, we throw an error - // this is a bit redundant (for good), since we're not sure if user validated the custom template with llama_chat_verify_template() - throw std::runtime_error("this custom template is not supported"); - } else { - // If the built-in template is not supported, we default to chatml - res = llama_chat_apply_template(nullptr, "chatml", chat.data(), chat.size(), add_ass, buf.data(), buf.size()); - fallback = true; - } + // if the custom "tmpl" is not supported, we throw an error + // this is a bit redundant (for good), since we're not sure if user validated the custom template with llama_chat_verify_template() + throw std::runtime_error("this custom template is not supported"); } // if it turns out that our buffer is too small, we resize it if ((size_t) res > buf.size()) { buf.resize(res); - res = llama_chat_apply_template( - fallback ? nullptr : model, - fallback ? "chatml" : ptr_tmpl, - chat.data(), chat.size(), add_ass, buf.data(), buf.size()); + res = llama_chat_apply_template(tmpl.source().c_str(), chat.data(), chat.size(), add_ass, buf.data(), buf.size()); } std::string formatted_chat(buf.data(), res); return formatted_chat; } -std::string common_chat_format_single(const struct llama_model * model, - const std::string & tmpl, +std::string common_chat_format_single( + const common_chat_template & tmpl, const std::vector & past_msg, const common_chat_msg & new_msg, - bool add_ass) { + bool add_ass, + bool use_jinja) { std::ostringstream ss; - auto fmt_past_msg = past_msg.empty() ? "" : common_chat_apply_template(model, tmpl, past_msg, false); + auto fmt_past_msg = past_msg.empty() ? "" : common_chat_apply_template(tmpl, past_msg, false, use_jinja); std::vector chat_new(past_msg); // if the past_msg ends with a newline, we must preserve it in the formatted version if (add_ass && !fmt_past_msg.empty() && fmt_past_msg.back() == '\n') { @@ -1690,21 +1847,74 @@ std::string common_chat_format_single(const struct llama_model * model, }; // format chat with new_msg chat_new.push_back(new_msg); - auto fmt_new_msg = common_chat_apply_template(model, tmpl, chat_new, add_ass); + auto fmt_new_msg = common_chat_apply_template(tmpl, chat_new, add_ass, use_jinja); // get the diff part ss << fmt_new_msg.substr(fmt_past_msg.size(), fmt_new_msg.size() - fmt_past_msg.size()); return ss.str(); } -std::string common_chat_format_example(const struct llama_model * model, - const std::string & tmpl) { +std::string common_chat_format_example(const common_chat_template & tmpl, bool use_jinja) { std::vector msgs = { {"system", "You are a helpful assistant"}, {"user", "Hello"}, {"assistant", "Hi there"}, {"user", "How are you?"}, }; - return common_chat_apply_template(model, tmpl, msgs, true); + return common_chat_apply_template(tmpl, msgs, true, use_jinja); +} + +common_chat_templates common_chat_templates_from_model(const struct llama_model * model, const std::string & chat_template_override) +{ + auto vocab = llama_model_get_vocab(model); + std::string default_template_src = chat_template_override; + std::string template_tool_use_src = chat_template_override; + bool has_explicit_template = !chat_template_override.empty(); + if (chat_template_override.empty()) { + auto str = llama_model_chat_template(model, /* name */ nullptr); + if (str) { + default_template_src = str; + has_explicit_template = true; + } + str = llama_model_chat_template(model, /* name */ "tool_use"); + if (str) { + template_tool_use_src = str; + has_explicit_template = true; + } + } + if (default_template_src.empty() || default_template_src == "chatml") { + if (!template_tool_use_src.empty()) { + default_template_src = template_tool_use_src; + } else { + default_template_src = R"( + {%- for message in messages -%} + {{- "<|im_start|>" + message.role + "\n" + message.content + "<|im_end|>\n" -}} + {%- endfor -%} + {%- if add_generation_prompt -%} + {{- "<|im_start|>assistant\n" -}} + {%- endif -%} + )"; + } + } + const auto get_token = [&](llama_token token, const char * name, const char * jinja_variable_name) { + if (token == LLAMA_TOKEN_NULL) { + if (default_template_src.find(jinja_variable_name) != std::string::npos + || template_tool_use_src.find(jinja_variable_name) != std::string::npos) { + LOG_WRN("%s: warning: vocab does not have a %s token, jinja template won't work as intended.\n", __func__, name); + } + return std::string(); + } else { + return common_token_to_piece(vocab, token, true); + } + }; + auto token_bos = get_token(llama_vocab_bos(vocab), "BOS", "bos_token"); + auto token_eos = get_token(llama_vocab_eos(vocab), "EOS", "eos_token"); + return { + has_explicit_template, + std::make_unique(default_template_src, token_bos, token_eos), + template_tool_use_src.empty() + ? nullptr + : std::make_unique(template_tool_use_src, token_bos, token_eos) + }; } // diff --git a/common/common.h b/common/common.h index 0d452cf0f..571260372 100644 --- a/common/common.h +++ b/common/common.h @@ -24,11 +24,11 @@ #define DEFAULT_MODEL_PATH "models/7B/ggml-model-f16.gguf" -struct common_lora_adapter_info { +struct common_adapter_lora_info { std::string path; float scale; - struct llama_lora_adapter * ptr; + struct llama_adapter_lora * ptr; }; using llama_tokens = std::vector; @@ -103,6 +103,12 @@ enum dimre_method { DIMRE_METHOD_MEAN, }; +enum common_conversation_mode { + COMMON_CONVERSATION_MODE_DISABLED = 0, + COMMON_CONVERSATION_MODE_ENABLED = 1, + COMMON_CONVERSATION_MODE_AUTO = 2, +}; + // sampling parameters struct common_params_sampling { uint32_t seed = LLAMA_DEFAULT_SEED; // the seed used to initialize llama_sampler @@ -169,7 +175,11 @@ struct common_params_speculative { struct cpu_params cpuparams; struct cpu_params cpuparams_batch; - std::string model = ""; // draft model for speculative decoding // NOLINT + std::string hf_repo = ""; // HF repo // NOLINT + std::string hf_file = ""; // HF file // NOLINT + + std::string model = ""; // draft model for speculative decoding // NOLINT + std::string model_url = ""; // model url to download // NOLINT }; struct common_params_vocoder { @@ -178,6 +188,8 @@ struct common_params_vocoder { std::string model = ""; // model path // NOLINT std::string model_url = ""; // model url to download // NOLINT + + bool use_guide_tokens = false; // enable guide tokens to improve TTS accuracy // NOLINT }; struct common_params { @@ -240,14 +252,13 @@ struct common_params { std::string lookup_cache_static = ""; // path of static ngram cache file for lookup decoding // NOLINT std::string lookup_cache_dynamic = ""; // path of dynamic ngram cache file for lookup decoding // NOLINT std::string logits_file = ""; // file for saving *all* logits // NOLINT - std::string rpc_servers = ""; // comma separated list of RPC servers // NOLINT std::vector in_files; // all input files std::vector antiprompt; // strings upon which more user input is prompted (a.k.a. reverse prompts) std::vector kv_overrides; - bool lora_init_without_apply = false; // only load lora to memory, but do not apply it to ctx (user can manually apply lora later using llama_lora_adapter_apply) - std::vector lora_adapters; // lora adapter path with user defined scale + bool lora_init_without_apply = false; // only load lora to memory, but do not apply it to ctx (user can manually apply lora later using llama_adapter_lora_apply) + std::vector lora_adapters; // lora adapter path with user defined scale std::vector control_vectors; // control vector with user defined scale @@ -275,7 +286,6 @@ struct common_params { bool special = false; // enable special token output bool interactive = false; // interactive mode bool interactive_first = false; // wait for user input immediately - bool conversation = false; // conversation mode (does not print special tokens and suffix/prefix) bool prompt_cache_all = false; // save user input and generations to prompt cache bool prompt_cache_ro = false; // open the prompt cache read-only and do not update it @@ -301,6 +311,8 @@ struct common_params { ggml_type cache_type_k = GGML_TYPE_F16; // KV cache data type for the K ggml_type cache_type_v = GGML_TYPE_F16; // KV cache data type for the V + common_conversation_mode conversation_mode = COMMON_CONVERSATION_MODE_AUTO; + // multimodal models (see examples/llava) std::string mmproj = ""; // path to multimodal projector // NOLINT std::vector image; // path to image file(s) @@ -322,6 +334,7 @@ struct common_params { std::string hostname = "127.0.0.1"; std::string public_path = ""; // NOLINT std::string chat_template = ""; // NOLINT + bool use_jinja = false; // NOLINT bool enable_chat_template = true; std::vector api_keys; @@ -416,6 +429,10 @@ std::string string_format(const char * fmt, ...); std::string string_strip(const std::string & str); std::string string_get_sortable_timestamp(); +std::string string_join(const std::vector & values, const std::string & separator); +std::vector string_split(const std::string & str, const std::string & delimiter); +std::string string_repeat(const std::string & str, size_t n); + void string_replace_all(std::string & s, const std::string & search, const std::string & replace); template @@ -454,6 +471,11 @@ static bool string_starts_with(const std::string & str, return str.rfind(prefix, 0) == 0; } +static bool string_ends_with(const std::string & str, + const std::string & suffix) { // While we wait for C++20's std::string::ends_with... + return str.size() >= suffix.size() && str.compare(str.size()-suffix.size(), suffix.size(), suffix) == 0; +} + bool string_parse_kv_override(const char * data, std::vector & overrides); void string_process_escapes(std::string & input); @@ -481,7 +503,7 @@ struct common_init_result { llama_model_ptr model; llama_context_ptr context; - std::vector lora; + std::vector lora; }; struct common_init_result common_init_from_params(common_params & params); @@ -495,6 +517,7 @@ struct llama_model * common_load_model_from_url( const std::string & local_path, const std::string & hf_token, const struct llama_model_params & params); + struct llama_model * common_load_model_from_hf( const std::string & repo, const std::string & remote_path, @@ -502,8 +525,12 @@ struct llama_model * common_load_model_from_hf( const std::string & hf_token, const struct llama_model_params & params); +std::pair common_get_hf_file( + const std::string & hf_repo_with_tag, + const std::string & hf_token); + // clear LoRA adapters from context, then apply new list of adapters -void common_lora_adapters_apply(struct llama_context * ctx, std::vector & lora); +void common_set_adapter_lora(struct llama_context * ctx, std::vector & lora); // // Batch utils @@ -541,7 +568,7 @@ std::vector common_tokenize( bool parse_special = false); std::vector common_tokenize( - const struct llama_model * model, + const struct llama_vocab * vocab, const std::string & text, bool add_special, bool parse_special = false); @@ -553,11 +580,21 @@ std::string common_token_to_piece( llama_token token, bool special = true); +std::string common_token_to_piece( + const struct llama_vocab * vocab, + llama_token token, + bool special = true); + // detokenizes a vector of tokens into a string // should work similar to Python's `tokenizer.decode` // optionally renders special/control tokens std::string common_detokenize( - llama_context * ctx, + const struct llama_context * ctx, + const std::vector & tokens, + bool special = true); + +std::string common_detokenize( + const struct llama_vocab * vocab, const std::vector & tokens, bool special = true); @@ -571,30 +608,43 @@ struct common_chat_msg { std::string content; }; -// Get the built-in chat template for the model. Return empty string if not present. -std::string common_get_builtin_chat_template(const struct llama_model * model); - // Check if the template supplied via "--chat-template" is supported or not. Returns true if it's valid -bool common_chat_verify_template(const std::string & tmpl); +bool common_chat_verify_template(const std::string & tmpl, bool use_jinja); + +namespace minja { + class chat_template; +} + +typedef minja::chat_template common_chat_template; + +struct common_chat_templates { + bool has_explicit_template; // Model had builtin template or template overridde was specified. + std::unique_ptr template_default; // always set (defaults to chatml) + std::unique_ptr template_tool_use; +}; // CPP wrapper for llama_chat_apply_template // If the built-in template is not supported, we default to chatml // If the custom "tmpl" is not supported, we throw an error -std::string common_chat_apply_template(const struct llama_model * model, - const std::string & tmpl, +std::string common_chat_apply_template( + const common_chat_template & tmpl, const std::vector & chat, - bool add_ass); + bool add_ass, + bool use_jinja); // Format single message, while taking into account the position of that message in chat history -std::string common_chat_format_single(const struct llama_model * model, - const std::string & tmpl, +std::string common_chat_format_single( + const common_chat_template & tmpl, const std::vector & past_msg, const common_chat_msg & new_msg, - bool add_ass); + bool add_ass, + bool use_jinja); // Returns an example of formatted chat -std::string common_chat_format_example(const struct llama_model * model, - const std::string & tmpl); +std::string common_chat_format_example( + const common_chat_template & tmpl, bool use_jinja); + +common_chat_templates common_chat_templates_from_model(const struct llama_model * model, const std::string & chat_template_override); // // KV cache utils diff --git a/common/json-schema-to-grammar.cpp b/common/json-schema-to-grammar.cpp index dadc18c8b..4d426b6bd 100644 --- a/common/json-schema-to-grammar.cpp +++ b/common/json-schema-to-grammar.cpp @@ -1,4 +1,6 @@ #include "json-schema-to-grammar.h" +#include "common.h" + #include #include #include @@ -11,11 +13,6 @@ using json = nlohmann::ordered_json; -template -static std::string join(Iterator begin, Iterator end, const std::string & separator); - -static std::string repeat(const std::string & str, size_t n); - static std::string build_repetition(const std::string & item_rule, int min_items, int max_items, const std::string & separator_rule = "") { auto has_max = max_items != std::numeric_limits::max(); @@ -128,8 +125,8 @@ static void _build_min_max_int(int min_value, int max_value, std::stringstream & if (sub_len > 0) { auto from_sub = from.substr(i + 1); auto to_sub = to.substr(i + 1); - auto sub_zeros = repeat("0", sub_len); - auto sub_nines = repeat("9", sub_len); + auto sub_zeros = string_repeat("0", sub_len); + auto sub_nines = string_repeat("9", sub_len); auto to_reached = false; out << "("; @@ -188,8 +185,8 @@ static void _build_min_max_int(int min_value, int max_value, std::stringstream & auto max_digits = max_s.length(); for (auto digits = min_digits; digits < max_digits; digits++) { - uniform_range(min_s, repeat("9", digits)); - min_s = "1" + repeat("0", digits); + uniform_range(min_s, string_repeat("9", digits)); + min_s = "1" + string_repeat("0", digits); out << " | "; } uniform_range(min_s, max_s); @@ -318,49 +315,6 @@ std::unordered_map GRAMMAR_LITERAL_ESCAPES = { std::unordered_set NON_LITERAL_SET = {'|', '.', '(', ')', '[', ']', '{', '}', '*', '+', '?'}; std::unordered_set ESCAPED_IN_REGEXPS_BUT_NOT_IN_LITERALS = {'^', '$', '.', '[', ']', '(', ')', '|', '{', '}', '*', '+', '?'}; -template -std::string join(Iterator begin, Iterator end, const std::string & separator) { - std::ostringstream result; - if (begin != end) { - result << *begin; - for (Iterator it = begin + 1; it != end; ++it) { - result << separator << *it; - } - } - return result.str(); -} - -static std::vector split(const std::string & str, const std::string & delimiter) { - std::vector tokens; - size_t start = 0; - size_t end = str.find(delimiter); - - while (end != std::string::npos) { - tokens.push_back(str.substr(start, end - start)); - start = end + delimiter.length(); - end = str.find(delimiter, start); - } - - tokens.push_back(str.substr(start)); - - return tokens; -} - -static std::string repeat(const std::string & str, size_t n) { - if (n == 0) { - return ""; - } - - std::string result; - result.reserve(str.length() * n); - - for (size_t i = 0; i < n; ++i) { - result += str; - } - - return result; -} - static std::string replacePattern(const std::string & input, const std::regex & regex, const std::function & replacement) { std::smatch match; std::string result; @@ -389,6 +343,7 @@ static std::string format_literal(const std::string & literal) { class SchemaConverter { private: + friend std::string build_grammar(const std::function & cb); std::function _fetch_json; bool _dotall; std::map _rules; @@ -418,7 +373,7 @@ private: for (size_t i = 0; i < alt_schemas.size(); i++) { rules.push_back(visit(alt_schemas[i], name + (name.empty() ? "alternative-" : "-") + std::to_string(i))); } - return join(rules.begin(), rules.end(), " | "); + return string_join(rules, " | "); } std::string _visit_pattern(const std::string & pattern, const std::string & name) { @@ -481,7 +436,7 @@ private: for (const auto & item : ret) { results.push_back(to_rule(item)); } - return std::make_pair(join(results.begin(), results.end(), " "), false); + return std::make_pair(string_join(results, " "), false); }; while (i < length) { @@ -539,7 +494,7 @@ private: } curly_brackets += '}'; i++; - auto nums = split(curly_brackets.substr(1, curly_brackets.length() - 2), ","); + auto nums = string_split(curly_brackets.substr(1, curly_brackets.length() - 2), ","); int min_times = 0; int max_times = std::numeric_limits::max(); try { @@ -854,7 +809,7 @@ public: return; } std::string pointer = ref.substr(ref.find('#') + 1); - std::vector tokens = split(pointer, "/"); + std::vector tokens = string_split(pointer, "/"); for (size_t i = 1; i < tokens.size(); ++i) { std::string sel = tokens[i]; if (target.is_null() || !target.contains(sel)) { @@ -905,7 +860,7 @@ public: for (const auto & v : schema["enum"]) { enum_values.push_back(_generate_constant_rule(v)); } - return _add_rule(rule_name, "(" + join(enum_values.begin(), enum_values.end(), " | ") + ") space"); + return _add_rule(rule_name, "(" + string_join(enum_values, " | ") + ") space"); } else if ((schema_type.is_null() || schema_type == "object") && (schema.contains("properties") || (schema.contains("additionalProperties") && schema["additionalProperties"] != true))) { @@ -1019,10 +974,10 @@ public: void check_errors() { if (!_errors.empty()) { - throw std::runtime_error("JSON schema conversion failed:\n" + join(_errors.begin(), _errors.end(), "\n")); + throw std::runtime_error("JSON schema conversion failed:\n" + string_join(_errors, "\n")); } if (!_warnings.empty()) { - fprintf(stderr, "WARNING: JSON schema conversion was incomplete: %s\n", join(_warnings.begin(), _warnings.end(), "; ").c_str()); + fprintf(stderr, "WARNING: JSON schema conversion was incomplete: %s\n", string_join(_warnings, "; ").c_str()); } } @@ -1036,10 +991,27 @@ public: }; std::string json_schema_to_grammar(const json & schema) { - SchemaConverter converter([](const std::string &) { return json::object(); }, /* dotall= */ false); - auto copy = schema; - converter.resolve_refs(copy, "input"); - converter.visit(copy, ""); + return build_grammar([&](const llama_grammar_builder & callbacks) { + auto copy = schema; + callbacks.resolve_refs(copy); + callbacks.add_schema("", copy); + }); +} + +std::string build_grammar(const std::function & cb) { + SchemaConverter converter([&](const std::string &) { return json(); }, /* dotall= */ false); + llama_grammar_builder builder { + /* .add_rule = */ [&](const std::string & name, const std::string & rule) { + return converter._add_rule(name, rule); + }, + /* .add_schema = */ [&](const std::string & name, const nlohmann::ordered_json & schema) { + return converter.visit(schema, name == "root" ? "" : name); + }, + /* .resolve_refs = */ [&](nlohmann::ordered_json & schema) { + converter.resolve_refs(schema, ""); + } + }; + cb(builder); converter.check_errors(); return converter.format_grammar(); } diff --git a/common/json-schema-to-grammar.h b/common/json-schema-to-grammar.h index 41623b346..4f43ab3a5 100644 --- a/common/json-schema-to-grammar.h +++ b/common/json-schema-to-grammar.h @@ -5,4 +5,12 @@ #define JSON_ASSERT GGML_ASSERT #include "json.hpp" -std::string json_schema_to_grammar(const nlohmann::ordered_json& schema); +std::string json_schema_to_grammar(const nlohmann::ordered_json & schema); + +struct llama_grammar_builder { + std::function add_rule; + std::function add_schema; + std::function resolve_refs; +}; + +std::string build_grammar(const std::function & cb); diff --git a/common/minja.hpp b/common/minja.hpp new file mode 100644 index 000000000..80bdd4b41 --- /dev/null +++ b/common/minja.hpp @@ -0,0 +1,2812 @@ +/* + Copyright 2024 Google LLC + + Use of this source code is governed by an MIT-style + license that can be found in the LICENSE file or at + https://opensource.org/licenses/MIT. +*/ +// SPDX-License-Identifier: MIT +#pragma once + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +using json = nlohmann::ordered_json; + +namespace minja { + +class Context; + +struct Options { + bool trim_blocks; // removes the first newline after a block + bool lstrip_blocks; // removes leading whitespace on the line of the block + bool keep_trailing_newline; // don't remove last newline +}; + +struct ArgumentsValue; + +inline std::string normalize_newlines(const std::string & s) { +#ifdef _WIN32 + static const std::regex nl_regex("\r\n"); + return std::regex_replace(s, nl_regex, "\n"); +#else + return s; +#endif +} + +/* Values that behave roughly like in Python. */ +class Value : public std::enable_shared_from_this { +public: + using CallableType = std::function &, ArgumentsValue &)>; + using FilterType = std::function &, ArgumentsValue &)>; + +private: + using ObjectType = nlohmann::ordered_map; // Only contains primitive keys + using ArrayType = std::vector; + + std::shared_ptr array_; + std::shared_ptr object_; + std::shared_ptr callable_; + json primitive_; + + Value(const std::shared_ptr & array) : array_(array) {} + Value(const std::shared_ptr & object) : object_(object) {} + Value(const std::shared_ptr & callable) : object_(std::make_shared()), callable_(callable) {} + + /* Python-style string repr */ + static void dump_string(const json & primitive, std::ostringstream & out, char string_quote = '\'') { + if (!primitive.is_string()) throw std::runtime_error("Value is not a string: " + primitive.dump()); + auto s = primitive.dump(); + if (string_quote == '"' || s.find('\'') != std::string::npos) { + out << s; + return; + } + // Reuse json dump, just changing string quotes + out << string_quote; + for (size_t i = 1, n = s.size() - 1; i < n; ++i) { + if (s[i] == '\\' && s[i + 1] == '"') { + out << '"'; + i++; + } else if (s[i] == string_quote) { + out << '\\' << string_quote; + } else { + out << s[i]; + } + } + out << string_quote; + } + void dump(std::ostringstream & out, int indent = -1, int level = 0, bool to_json = false) const { + auto print_indent = [&](int level) { + if (indent > 0) { + out << "\n"; + for (int i = 0, n = level * indent; i < n; ++i) out << ' '; + } + }; + auto print_sub_sep = [&]() { + out << ','; + if (indent < 0) out << ' '; + else print_indent(level + 1); + }; + + auto string_quote = to_json ? '"' : '\''; + + if (is_null()) out << "null"; + else if (array_) { + out << "["; + print_indent(level + 1); + for (size_t i = 0; i < array_->size(); ++i) { + if (i) print_sub_sep(); + (*array_)[i].dump(out, indent, level + 1, to_json); + } + print_indent(level); + out << "]"; + } else if (object_) { + out << "{"; + print_indent(level + 1); + for (auto begin = object_->begin(), it = begin; it != object_->end(); ++it) { + if (it != begin) print_sub_sep(); + if (it->first.is_string()) { + dump_string(it->first, out, string_quote); + } else { + out << string_quote << it->first.dump() << string_quote; + } + out << ": "; + it->second.dump(out, indent, level + 1, to_json); + } + print_indent(level); + out << "}"; + } else if (callable_) { + throw std::runtime_error("Cannot dump callable to JSON"); + } else if (is_boolean() && !to_json) { + out << (this->to_bool() ? "True" : "False"); + } else if (is_string() && !to_json) { + dump_string(primitive_, out, string_quote); + } else { + out << primitive_.dump(); + } + } + +public: + Value() {} + Value(const bool& v) : primitive_(v) {} + Value(const int64_t & v) : primitive_(v) {} + Value(const double& v) : primitive_(v) {} + Value(const std::nullptr_t &) {} + Value(const std::string & v) : primitive_(v) {} + Value(const char * v) : primitive_(std::string(v)) {} + + Value(const json & v) { + if (v.is_object()) { + auto object = std::make_shared(); + for (auto it = v.begin(); it != v.end(); ++it) { + (*object)[it.key()] = it.value(); + } + object_ = std::move(object); + } else if (v.is_array()) { + auto array = std::make_shared(); + for (const auto& item : v) { + array->push_back(Value(item)); + } + array_ = array; + } else { + primitive_ = v; + } + } + + std::vector keys() { + if (!object_) throw std::runtime_error("Value is not an object: " + dump()); + std::vector res; + for (const auto& item : *object_) { + res.push_back(item.first); + } + return res; + } + + size_t size() const { + if (is_object()) return object_->size(); + if (is_array()) return array_->size(); + if (is_string()) return primitive_.get().length(); + throw std::runtime_error("Value is not an array or object: " + dump()); + } + + static Value array(const std::vector values = {}) { + auto array = std::make_shared(); + for (const auto& item : values) { + array->push_back(item); + } + return Value(array); + } + static Value object(const std::shared_ptr object = std::make_shared()) { + return Value(object); + } + static Value callable(const CallableType & callable) { + return Value(std::make_shared(callable)); + } + + void insert(size_t index, const Value& v) { + if (!array_) + throw std::runtime_error("Value is not an array: " + dump()); + array_->insert(array_->begin() + index, v); + } + void push_back(const Value& v) { + if (!array_) + throw std::runtime_error("Value is not an array: " + dump()); + array_->push_back(v); + } + Value pop(const Value& index) { + if (is_array()) { + if (array_->empty()) + throw std::runtime_error("pop from empty list"); + if (index.is_null()) { + auto ret = array_->back(); + array_->pop_back(); + return ret; + } else if (!index.is_number_integer()) { + throw std::runtime_error("pop index must be an integer: " + index.dump()); + } else { + auto i = index.get(); + if (i < 0 || i >= static_cast(array_->size())) + throw std::runtime_error("pop index out of range: " + index.dump()); + auto it = array_->begin() + (i < 0 ? array_->size() + i : i); + auto ret = *it; + array_->erase(it); + return ret; + } + } else if (is_object()) { + if (!index.is_hashable()) + throw std::runtime_error("Unashable type: " + index.dump()); + auto it = object_->find(index.primitive_); + if (it == object_->end()) + throw std::runtime_error("Key not found: " + index.dump()); + auto ret = it->second; + object_->erase(it); + return ret; + } else { + throw std::runtime_error("Value is not an array or object: " + dump()); + } + } + Value get(const Value& key) { + if (array_) { + if (!key.is_number_integer()) { + return Value(); + } + auto index = key.get(); + return array_->at(index < 0 ? array_->size() + index : index); + } else if (object_) { + if (!key.is_hashable()) throw std::runtime_error("Unashable type: " + dump()); + auto it = object_->find(key.primitive_); + if (it == object_->end()) return Value(); + return it->second; + } + return Value(); + } + void set(const Value& key, const Value& value) { + if (!object_) throw std::runtime_error("Value is not an object: " + dump()); + if (!key.is_hashable()) throw std::runtime_error("Unashable type: " + dump()); + (*object_)[key.primitive_] = value; + } + Value call(const std::shared_ptr & context, ArgumentsValue & args) const { + if (!callable_) throw std::runtime_error("Value is not callable: " + dump()); + return (*callable_)(context, args); + } + + bool is_object() const { return !!object_; } + bool is_array() const { return !!array_; } + bool is_callable() const { return !!callable_; } + bool is_null() const { return !object_ && !array_ && primitive_.is_null() && !callable_; } + bool is_boolean() const { return primitive_.is_boolean(); } + bool is_number_integer() const { return primitive_.is_number_integer(); } + bool is_number_float() const { return primitive_.is_number_float(); } + bool is_number() const { return primitive_.is_number(); } + bool is_string() const { return primitive_.is_string(); } + bool is_iterable() const { return is_array() || is_object() || is_string(); } + + bool is_primitive() const { return !array_ && !object_ && !callable_; } + bool is_hashable() const { return is_primitive(); } + + bool empty() const { + if (is_null()) + throw std::runtime_error("Undefined value or reference"); + if (is_string()) return primitive_.empty(); + if (is_array()) return array_->empty(); + if (is_object()) return object_->empty(); + return false; + } + + void for_each(const std::function & callback) const { + if (is_null()) + throw std::runtime_error("Undefined value or reference"); + if (array_) { + for (auto& item : *array_) { + callback(item); + } + } else if (object_) { + for (auto & item : *object_) { + Value key(item.first); + callback(key); + } + } else if (is_string()) { + for (char c : primitive_.get()) { + auto val = Value(std::string(1, c)); + callback(val); + } + } else { + throw std::runtime_error("Value is not iterable: " + dump()); + } + } + + bool to_bool() const { + if (is_null()) return false; + if (is_boolean()) return get(); + if (is_number()) return get() != 0; + if (is_string()) return !get().empty(); + if (is_array()) return !empty(); + return true; + } + + int64_t to_int() const { + if (is_null()) return 0; + if (is_boolean()) return get() ? 1 : 0; + if (is_number()) return static_cast(get()); + if (is_string()) { + try { + return std::stol(get()); + } catch (const std::exception &) { + return 0; + } + } + return 0; + } + + bool operator<(const Value & other) const { + if (is_null()) + throw std::runtime_error("Undefined value or reference"); + if (is_number() && other.is_number()) return get() < other.get(); + if (is_string() && other.is_string()) return get() < other.get(); + throw std::runtime_error("Cannot compare values: " + dump() + " < " + other.dump()); + } + bool operator>=(const Value & other) const { return !(*this < other); } + + bool operator>(const Value & other) const { + if (is_null()) + throw std::runtime_error("Undefined value or reference"); + if (is_number() && other.is_number()) return get() > other.get(); + if (is_string() && other.is_string()) return get() > other.get(); + throw std::runtime_error("Cannot compare values: " + dump() + " > " + other.dump()); + } + bool operator<=(const Value & other) const { return !(*this > other); } + + bool operator==(const Value & other) const { + if (callable_ || other.callable_) { + if (callable_.get() != other.callable_.get()) return false; + } + if (array_) { + if (!other.array_) return false; + if (array_->size() != other.array_->size()) return false; + for (size_t i = 0; i < array_->size(); ++i) { + if (!(*array_)[i].to_bool() || !(*other.array_)[i].to_bool() || (*array_)[i] != (*other.array_)[i]) return false; + } + return true; + } else if (object_) { + if (!other.object_) return false; + if (object_->size() != other.object_->size()) return false; + for (const auto& item : *object_) { + if (!item.second.to_bool() || !other.object_->count(item.first) || item.second != other.object_->at(item.first)) return false; + } + return true; + } else { + return primitive_ == other.primitive_; + } + } + bool operator!=(const Value & other) const { return !(*this == other); } + + bool contains(const char * key) const { return contains(std::string(key)); } + bool contains(const std::string & key) const { + if (array_) { + return false; + } else if (object_) { + return object_->find(key) != object_->end(); + } else { + throw std::runtime_error("contains can only be called on arrays and objects: " + dump()); + } + } + bool contains(const Value & value) const { + if (is_null()) + throw std::runtime_error("Undefined value or reference"); + if (array_) { + for (const auto& item : *array_) { + if (item.to_bool() && item == value) return true; + } + return false; + } else if (object_) { + if (!value.is_hashable()) throw std::runtime_error("Unashable type: " + value.dump()); + return object_->find(value.primitive_) != object_->end(); + } else { + throw std::runtime_error("contains can only be called on arrays and objects: " + dump()); + } + } + void erase(size_t index) { + if (!array_) throw std::runtime_error("Value is not an array: " + dump()); + array_->erase(array_->begin() + index); + } + void erase(const std::string & key) { + if (!object_) throw std::runtime_error("Value is not an object: " + dump()); + object_->erase(key); + } + const Value& at(const Value & index) const { + return const_cast(this)->at(index); + } + Value& at(const Value & index) { + if (!index.is_hashable()) throw std::runtime_error("Unashable type: " + dump()); + if (is_array()) return array_->at(index.get()); + if (is_object()) return object_->at(index.primitive_); + throw std::runtime_error("Value is not an array or object: " + dump()); + } + const Value& at(size_t index) const { + return const_cast(this)->at(index); + } + Value& at(size_t index) { + if (is_null()) + throw std::runtime_error("Undefined value or reference"); + if (is_array()) return array_->at(index); + if (is_object()) return object_->at(index); + throw std::runtime_error("Value is not an array or object: " + dump()); + } + + template + T get(const std::string & key, T default_value) const { + if (!contains(key)) return default_value; + return at(key).get(); + } + + template + T get() const { + if (is_primitive()) return primitive_.get(); + throw std::runtime_error("get not defined for this value type: " + dump()); + } + + std::string dump(int indent=-1, bool to_json=false) const { + std::ostringstream out; + dump(out, indent, 0, to_json); + return out.str(); + } + + Value operator-() const { + if (is_number_integer()) + return -get(); + else + return -get(); + } + std::string to_str() const { + if (is_string()) return get(); + if (is_number_integer()) return std::to_string(get()); + if (is_number_float()) return std::to_string(get()); + if (is_boolean()) return get() ? "True" : "False"; + if (is_null()) return "None"; + return dump(); + } + Value operator+(const Value& rhs) const { + if (is_string() || rhs.is_string()) { + return to_str() + rhs.to_str(); + } else if (is_number_integer() && rhs.is_number_integer()) { + return get() + rhs.get(); + } else if (is_array() && rhs.is_array()) { + auto res = Value::array(); + for (const auto& item : *array_) res.push_back(item); + for (const auto& item : *rhs.array_) res.push_back(item); + return res; + } else { + return get() + rhs.get(); + } + } + Value operator-(const Value& rhs) const { + if (is_number_integer() && rhs.is_number_integer()) + return get() - rhs.get(); + else + return get() - rhs.get(); + } + Value operator*(const Value& rhs) const { + if (is_string() && rhs.is_number_integer()) { + std::ostringstream out; + for (int64_t i = 0, n = rhs.get(); i < n; ++i) { + out << to_str(); + } + return out.str(); + } + else if (is_number_integer() && rhs.is_number_integer()) + return get() * rhs.get(); + else + return get() * rhs.get(); + } + Value operator/(const Value& rhs) const { + if (is_number_integer() && rhs.is_number_integer()) + return get() / rhs.get(); + else + return get() / rhs.get(); + } + Value operator%(const Value& rhs) const { + return get() % rhs.get(); + } +}; + +struct ArgumentsValue { + std::vector args; + std::vector> kwargs; + + bool has_named(const std::string & name) { + for (const auto & p : kwargs) { + if (p.first == name) return true; + } + return false; + } + + Value get_named(const std::string & name) { + for (const auto & [key, value] : kwargs) { + if (key == name) return value; + } + return Value(); + } + + bool empty() { + return args.empty() && kwargs.empty(); + } + + void expectArgs(const std::string & method_name, const std::pair & pos_count, const std::pair & kw_count) { + if (args.size() < pos_count.first || args.size() > pos_count.second || kwargs.size() < kw_count.first || kwargs.size() > kw_count.second) { + std::ostringstream out; + out << method_name << " must have between " << pos_count.first << " and " << pos_count.second << " positional arguments and between " << kw_count.first << " and " << kw_count.second << " keyword arguments"; + throw std::runtime_error(out.str()); + } + } +}; + +template <> +inline json Value::get() const { + if (is_primitive()) return primitive_; + if (is_null()) return json(); + if (array_) { + std::vector res; + for (const auto& item : *array_) { + res.push_back(item.get()); + } + return res; + } + if (object_) { + json res = json::object(); + for (const auto& [key, value] : *object_) { + if (key.is_string()) { + res[key.get()] = value.get(); + } else if (key.is_primitive()) { + res[key.dump()] = value.get(); + } else { + throw std::runtime_error("Invalid key type for conversion to JSON: " + key.dump()); + } + } + if (is_callable()) { + res["__callable__"] = true; + } + return res; + } + throw std::runtime_error("get not defined for this value type: " + dump()); +} + +} // namespace minja + +namespace std { + template <> + struct hash { + size_t operator()(const minja::Value & v) const { + if (!v.is_hashable()) + throw std::runtime_error("Unsupported type for hashing: " + v.dump()); + return std::hash()(v.get()); + } + }; +} // namespace std + +namespace minja { + +static std::string error_location_suffix(const std::string & source, size_t pos) { + auto get_line = [&](size_t line) { + auto start = source.begin(); + for (size_t i = 1; i < line; ++i) { + start = std::find(start, source.end(), '\n') + 1; + } + auto end = std::find(start, source.end(), '\n'); + return std::string(start, end); + }; + auto start = source.begin(); + auto end = source.end(); + auto it = start + pos; + auto line = std::count(start, it, '\n') + 1; + auto max_line = std::count(start, end, '\n') + 1; + auto col = pos - std::string(start, it).rfind('\n'); + std::ostringstream out; + out << " at row " << line << ", column " << col << ":\n"; + if (line > 1) out << get_line(line - 1) << "\n"; + out << get_line(line) << "\n"; + out << std::string(col - 1, ' ') << "^\n"; + if (line < max_line) out << get_line(line + 1) << "\n"; + + return out.str(); +} + +class Context : public std::enable_shared_from_this { + protected: + Value values_; + std::shared_ptr parent_; + public: + Context(Value && values, const std::shared_ptr & parent = nullptr) : values_(std::move(values)), parent_(parent) { + if (!values_.is_object()) throw std::runtime_error("Context values must be an object: " + values_.dump()); + } + virtual ~Context() {} + + static std::shared_ptr builtins(); + static std::shared_ptr make(Value && values, const std::shared_ptr & parent = builtins()); + + std::vector keys() { + return values_.keys(); + } + virtual Value get(const Value & key) { + if (values_.contains(key)) return values_.at(key); + if (parent_) return parent_->get(key); + return Value(); + } + virtual Value & at(const Value & key) { + if (values_.contains(key)) return values_.at(key); + if (parent_) return parent_->at(key); + throw std::runtime_error("Undefined variable: " + key.dump()); + } + virtual bool contains(const Value & key) { + if (values_.contains(key)) return true; + if (parent_) return parent_->contains(key); + return false; + } + virtual void set(const Value & key, Value & value) { + values_.set(key, value); + } +}; + +struct Location { + std::shared_ptr source; + size_t pos; +}; + +class Expression { +protected: + virtual Value do_evaluate(const std::shared_ptr & context) const = 0; +public: + using Parameters = std::vector>>; + + Location location; + + Expression(const Location & location) : location(location) {} + virtual ~Expression() = default; + + Value evaluate(const std::shared_ptr & context) const { + try { + return do_evaluate(context); + } catch (const std::exception & e) { + std::ostringstream out; + out << e.what(); + if (location.source) out << error_location_suffix(*location.source, location.pos); + throw std::runtime_error(out.str()); + } + } +}; + +class VariableExpr : public Expression { + std::string name; +public: + VariableExpr(const Location & location, const std::string& n) + : Expression(location), name(n) {} + std::string get_name() const { return name; } + Value do_evaluate(const std::shared_ptr & context) const override { + if (!context->contains(name)) { + return Value(); + } + return context->at(name); + } +}; + +static void destructuring_assign(const std::vector & var_names, const std::shared_ptr & context, Value& item) { + if (var_names.size() == 1) { + Value name(var_names[0]); + context->set(name, item); + } else { + if (!item.is_array() || item.size() != var_names.size()) { + throw std::runtime_error("Mismatched number of variables and items in destructuring assignment"); + } + for (size_t i = 0; i < var_names.size(); ++i) { + context->set(var_names[i], item.at(i)); + } + } +} + +enum SpaceHandling { Keep, Strip, StripSpaces, StripNewline }; + +class TemplateToken { +public: + enum class Type { Text, Expression, If, Else, Elif, EndIf, For, EndFor, Generation, EndGeneration, Set, EndSet, Comment, Macro, EndMacro, Filter, EndFilter }; + + static std::string typeToString(Type t) { + switch (t) { + case Type::Text: return "text"; + case Type::Expression: return "expression"; + case Type::If: return "if"; + case Type::Else: return "else"; + case Type::Elif: return "elif"; + case Type::EndIf: return "endif"; + case Type::For: return "for"; + case Type::EndFor: return "endfor"; + case Type::Set: return "set"; + case Type::EndSet: return "endset"; + case Type::Comment: return "comment"; + case Type::Macro: return "macro"; + case Type::EndMacro: return "endmacro"; + case Type::Filter: return "filter"; + case Type::EndFilter: return "endfilter"; + case Type::Generation: return "generation"; + case Type::EndGeneration: return "endgeneration"; + } + return "Unknown"; + } + + TemplateToken(Type type, const Location & location, SpaceHandling pre, SpaceHandling post) : type(type), location(location), pre_space(pre), post_space(post) {} + virtual ~TemplateToken() = default; + + Type type; + Location location; + SpaceHandling pre_space = SpaceHandling::Keep; + SpaceHandling post_space = SpaceHandling::Keep; +}; + +struct TextTemplateToken : public TemplateToken { + std::string text; + TextTemplateToken(const Location & location, SpaceHandling pre, SpaceHandling post, const std::string& t) : TemplateToken(Type::Text, location, pre, post), text(t) {} +}; + +struct ExpressionTemplateToken : public TemplateToken { + std::shared_ptr expr; + ExpressionTemplateToken(const Location & location, SpaceHandling pre, SpaceHandling post, std::shared_ptr && e) : TemplateToken(Type::Expression, location, pre, post), expr(std::move(e)) {} +}; + +struct IfTemplateToken : public TemplateToken { + std::shared_ptr condition; + IfTemplateToken(const Location & location, SpaceHandling pre, SpaceHandling post, std::shared_ptr && c) : TemplateToken(Type::If, location, pre, post), condition(std::move(c)) {} +}; + +struct ElifTemplateToken : public TemplateToken { + std::shared_ptr condition; + ElifTemplateToken(const Location & location, SpaceHandling pre, SpaceHandling post, std::shared_ptr && c) : TemplateToken(Type::Elif, location, pre, post), condition(std::move(c)) {} +}; + +struct ElseTemplateToken : public TemplateToken { + ElseTemplateToken(const Location & location, SpaceHandling pre, SpaceHandling post) : TemplateToken(Type::Else, location, pre, post) {} +}; + +struct EndIfTemplateToken : public TemplateToken { + EndIfTemplateToken(const Location & location, SpaceHandling pre, SpaceHandling post) : TemplateToken(Type::EndIf, location, pre, post) {} +}; + +struct MacroTemplateToken : public TemplateToken { + std::shared_ptr name; + Expression::Parameters params; + MacroTemplateToken(const Location & location, SpaceHandling pre, SpaceHandling post, std::shared_ptr && n, Expression::Parameters && p) + : TemplateToken(Type::Macro, location, pre, post), name(std::move(n)), params(std::move(p)) {} +}; + +struct EndMacroTemplateToken : public TemplateToken { + EndMacroTemplateToken(const Location & location, SpaceHandling pre, SpaceHandling post) : TemplateToken(Type::EndMacro, location, pre, post) {} +}; + +struct FilterTemplateToken : public TemplateToken { + std::shared_ptr filter; + FilterTemplateToken(const Location & location, SpaceHandling pre, SpaceHandling post, std::shared_ptr && filter) + : TemplateToken(Type::Filter, location, pre, post), filter(std::move(filter)) {} +}; + +struct EndFilterTemplateToken : public TemplateToken { + EndFilterTemplateToken(const Location & location, SpaceHandling pre, SpaceHandling post) : TemplateToken(Type::EndFilter, location, pre, post) {} +}; + +struct ForTemplateToken : public TemplateToken { + std::vector var_names; + std::shared_ptr iterable; + std::shared_ptr condition; + bool recursive; + ForTemplateToken(const Location & location, SpaceHandling pre, SpaceHandling post, const std::vector & vns, std::shared_ptr && iter, + std::shared_ptr && c, bool r) + : TemplateToken(Type::For, location, pre, post), var_names(vns), iterable(std::move(iter)), condition(std::move(c)), recursive(r) {} +}; + +struct EndForTemplateToken : public TemplateToken { + EndForTemplateToken(const Location & location, SpaceHandling pre, SpaceHandling post) : TemplateToken(Type::EndFor, location, pre, post) {} +}; + +struct GenerationTemplateToken : public TemplateToken { + GenerationTemplateToken(const Location & location, SpaceHandling pre, SpaceHandling post) : TemplateToken(Type::Generation, location, pre, post) {} +}; + +struct EndGenerationTemplateToken : public TemplateToken { + EndGenerationTemplateToken(const Location & location, SpaceHandling pre, SpaceHandling post) : TemplateToken(Type::EndGeneration, location, pre, post) {} +}; + +struct SetTemplateToken : public TemplateToken { + std::string ns; + std::vector var_names; + std::shared_ptr value; + SetTemplateToken(const Location & location, SpaceHandling pre, SpaceHandling post, const std::string & ns, const std::vector & vns, std::shared_ptr && v) + : TemplateToken(Type::Set, location, pre, post), ns(ns), var_names(vns), value(std::move(v)) {} +}; + +struct EndSetTemplateToken : public TemplateToken { + EndSetTemplateToken(const Location & location, SpaceHandling pre, SpaceHandling post) : TemplateToken(Type::EndSet, location, pre, post) {} +}; + +struct CommentTemplateToken : public TemplateToken { + std::string text; + CommentTemplateToken(const Location & location, SpaceHandling pre, SpaceHandling post, const std::string& t) : TemplateToken(Type::Comment, location, pre, post), text(t) {} +}; + +class TemplateNode { + Location location_; +protected: + virtual void do_render(std::ostringstream & out, const std::shared_ptr & context) const = 0; + +public: + TemplateNode(const Location & location) : location_(location) {} + void render(std::ostringstream & out, const std::shared_ptr & context) const { + try { + do_render(out, context); + } catch (const std::exception & e) { + std::ostringstream err; + err << e.what(); + if (location_.source) err << error_location_suffix(*location_.source, location_.pos); + throw std::runtime_error(err.str()); + } + } + const Location & location() const { return location_; } + virtual ~TemplateNode() = default; + std::string render(const std::shared_ptr & context) const { + std::ostringstream out; + render(out, context); + return out.str(); + } +}; + +class SequenceNode : public TemplateNode { + std::vector> children; +public: + SequenceNode(const Location & location, std::vector> && c) + : TemplateNode(location), children(std::move(c)) {} + void do_render(std::ostringstream & out, const std::shared_ptr & context) const override { + for (const auto& child : children) child->render(out, context); + } +}; + +class TextNode : public TemplateNode { + std::string text; +public: + TextNode(const Location & location, const std::string& t) : TemplateNode(location), text(t) {} + void do_render(std::ostringstream & out, const std::shared_ptr &) const override { + out << text; + } +}; + +class ExpressionNode : public TemplateNode { + std::shared_ptr expr; +public: + ExpressionNode(const Location & location, std::shared_ptr && e) : TemplateNode(location), expr(std::move(e)) {} + void do_render(std::ostringstream & out, const std::shared_ptr & context) const override { + if (!expr) throw std::runtime_error("ExpressionNode.expr is null"); + auto result = expr->evaluate(context); + if (result.is_string()) { + out << result.get(); + } else if (result.is_boolean()) { + out << (result.get() ? "True" : "False"); + } else if (!result.is_null()) { + out << result.dump(); + } + } +}; + +class IfNode : public TemplateNode { + std::vector, std::shared_ptr>> cascade; +public: + IfNode(const Location & location, std::vector, std::shared_ptr>> && c) + : TemplateNode(location), cascade(std::move(c)) {} + void do_render(std::ostringstream & out, const std::shared_ptr & context) const override { + for (const auto& branch : cascade) { + auto enter_branch = true; + if (branch.first) { + enter_branch = branch.first->evaluate(context).to_bool(); + } + if (enter_branch) { + if (!branch.second) throw std::runtime_error("IfNode.cascade.second is null"); + branch.second->render(out, context); + return; + } + } + } +}; + +class ForNode : public TemplateNode { + std::vector var_names; + std::shared_ptr iterable; + std::shared_ptr condition; + std::shared_ptr body; + bool recursive; + std::shared_ptr else_body; +public: + ForNode(const Location & location, std::vector && var_names, std::shared_ptr && iterable, + std::shared_ptr && condition, std::shared_ptr && body, bool recursive, std::shared_ptr && else_body) + : TemplateNode(location), var_names(var_names), iterable(std::move(iterable)), condition(std::move(condition)), body(std::move(body)), recursive(recursive), else_body(std::move(else_body)) {} + + void do_render(std::ostringstream & out, const std::shared_ptr & context) const override { + // https://jinja.palletsprojects.com/en/3.0.x/templates/#for + if (!iterable) throw std::runtime_error("ForNode.iterable is null"); + if (!body) throw std::runtime_error("ForNode.body is null"); + + auto iterable_value = iterable->evaluate(context); + Value::CallableType loop_function; + + std::function visit = [&](Value& iter) { + auto filtered_items = Value::array(); + if (!iter.is_null()) { + if (!iterable_value.is_iterable()) { + throw std::runtime_error("For loop iterable must be iterable: " + iterable_value.dump()); + } + iterable_value.for_each([&](Value & item) { + destructuring_assign(var_names, context, item); + if (!condition || condition->evaluate(context).to_bool()) { + filtered_items.push_back(item); + } + }); + } + if (filtered_items.empty()) { + if (else_body) { + else_body->render(out, context); + } + } else { + auto loop = recursive ? Value::callable(loop_function) : Value::object(); + loop.set("length", (int64_t) filtered_items.size()); + + size_t cycle_index = 0; + loop.set("cycle", Value::callable([&](const std::shared_ptr &, ArgumentsValue & args) { + if (args.args.empty() || !args.kwargs.empty()) { + throw std::runtime_error("cycle() expects at least 1 positional argument and no named arg"); + } + auto item = args.args[cycle_index]; + cycle_index = (cycle_index + 1) % args.args.size(); + return item; + })); + auto loop_context = Context::make(Value::object(), context); + loop_context->set("loop", loop); + for (size_t i = 0, n = filtered_items.size(); i < n; ++i) { + auto & item = filtered_items.at(i); + destructuring_assign(var_names, loop_context, item); + loop.set("index", (int64_t) i + 1); + loop.set("index0", (int64_t) i); + loop.set("revindex", (int64_t) (n - i)); + loop.set("revindex0", (int64_t) (n - i - 1)); + loop.set("length", (int64_t) n); + loop.set("first", i == 0); + loop.set("last", i == (n - 1)); + loop.set("previtem", i > 0 ? filtered_items.at(i - 1) : Value()); + loop.set("nextitem", i < n - 1 ? filtered_items.at(i + 1) : Value()); + body->render(out, loop_context); + } + } + }; + + if (recursive) { + loop_function = [&](const std::shared_ptr &, ArgumentsValue & args) { + if (args.args.size() != 1 || !args.kwargs.empty() || !args.args[0].is_array()) { + throw std::runtime_error("loop() expects exactly 1 positional iterable argument"); + } + auto & items = args.args[0]; + visit(items); + return Value(); + }; + } + + visit(iterable_value); + } +}; + +class MacroNode : public TemplateNode { + std::shared_ptr name; + Expression::Parameters params; + std::shared_ptr body; + std::unordered_map named_param_positions; +public: + MacroNode(const Location & location, std::shared_ptr && n, Expression::Parameters && p, std::shared_ptr && b) + : TemplateNode(location), name(std::move(n)), params(std::move(p)), body(std::move(b)) { + for (size_t i = 0; i < params.size(); ++i) { + const auto & name = params[i].first; + if (!name.empty()) { + named_param_positions[name] = i; + } + } + } + void do_render(std::ostringstream &, const std::shared_ptr & macro_context) const override { + if (!name) throw std::runtime_error("MacroNode.name is null"); + if (!body) throw std::runtime_error("MacroNode.body is null"); + auto callable = Value::callable([&](const std::shared_ptr & context, ArgumentsValue & args) { + auto call_context = macro_context; + std::vector param_set(params.size(), false); + for (size_t i = 0, n = args.args.size(); i < n; i++) { + auto & arg = args.args[i]; + if (i >= params.size()) throw std::runtime_error("Too many positional arguments for macro " + name->get_name()); + param_set[i] = true; + auto & param_name = params[i].first; + call_context->set(param_name, arg); + } + for (auto & [arg_name, value] : args.kwargs) { + auto it = named_param_positions.find(arg_name); + if (it == named_param_positions.end()) throw std::runtime_error("Unknown parameter name for macro " + name->get_name() + ": " + arg_name); + + call_context->set(arg_name, value); + param_set[it->second] = true; + } + // Set default values for parameters that were not passed + for (size_t i = 0, n = params.size(); i < n; i++) { + if (!param_set[i] && params[i].second != nullptr) { + auto val = params[i].second->evaluate(context); + call_context->set(params[i].first, val); + } + } + return body->render(call_context); + }); + macro_context->set(name->get_name(), callable); + } +}; + +class FilterNode : public TemplateNode { + std::shared_ptr filter; + std::shared_ptr body; + +public: + FilterNode(const Location & location, std::shared_ptr && f, std::shared_ptr && b) + : TemplateNode(location), filter(std::move(f)), body(std::move(b)) {} + + void do_render(std::ostringstream & out, const std::shared_ptr & context) const override { + if (!filter) throw std::runtime_error("FilterNode.filter is null"); + if (!body) throw std::runtime_error("FilterNode.body is null"); + auto filter_value = filter->evaluate(context); + if (!filter_value.is_callable()) { + throw std::runtime_error("Filter must be a callable: " + filter_value.dump()); + } + std::string rendered_body = body->render(context); + + ArgumentsValue filter_args = {{Value(rendered_body)}, {}}; + auto result = filter_value.call(context, filter_args); + out << result.to_str(); + } +}; + +class SetNode : public TemplateNode { + std::string ns; + std::vector var_names; + std::shared_ptr value; +public: + SetNode(const Location & location, const std::string & ns, const std::vector & vns, std::shared_ptr && v) + : TemplateNode(location), ns(ns), var_names(vns), value(std::move(v)) {} + void do_render(std::ostringstream &, const std::shared_ptr & context) const override { + if (!value) throw std::runtime_error("SetNode.value is null"); + if (!ns.empty()) { + if (var_names.size() != 1) { + throw std::runtime_error("Namespaced set only supports a single variable name"); + } + auto & name = var_names[0]; + auto ns_value = context->get(ns); + if (!ns_value.is_object()) throw std::runtime_error("Namespace '" + ns + "' is not an object"); + ns_value.set(name, this->value->evaluate(context)); + } else { + auto val = value->evaluate(context); + destructuring_assign(var_names, context, val); + } + } +}; + +class SetTemplateNode : public TemplateNode { + std::string name; + std::shared_ptr template_value; +public: + SetTemplateNode(const Location & location, const std::string & name, std::shared_ptr && tv) + : TemplateNode(location), name(name), template_value(std::move(tv)) {} + void do_render(std::ostringstream &, const std::shared_ptr & context) const override { + if (!template_value) throw std::runtime_error("SetTemplateNode.template_value is null"); + Value value { template_value->render(context) }; + context->set(name, value); + } +}; + +class IfExpr : public Expression { + std::shared_ptr condition; + std::shared_ptr then_expr; + std::shared_ptr else_expr; +public: + IfExpr(const Location & location, std::shared_ptr && c, std::shared_ptr && t, std::shared_ptr && e) + : Expression(location), condition(std::move(c)), then_expr(std::move(t)), else_expr(std::move(e)) {} + Value do_evaluate(const std::shared_ptr & context) const override { + if (!condition) throw std::runtime_error("IfExpr.condition is null"); + if (!then_expr) throw std::runtime_error("IfExpr.then_expr is null"); + if (condition->evaluate(context).to_bool()) { + return then_expr->evaluate(context); + } + if (else_expr) { + return else_expr->evaluate(context); + } + return nullptr; + } +}; + +class LiteralExpr : public Expression { + Value value; +public: + LiteralExpr(const Location & location, const Value& v) + : Expression(location), value(v) {} + Value do_evaluate(const std::shared_ptr &) const override { return value; } +}; + +class ArrayExpr : public Expression { + std::vector> elements; +public: + ArrayExpr(const Location & location, std::vector> && e) + : Expression(location), elements(std::move(e)) {} + Value do_evaluate(const std::shared_ptr & context) const override { + auto result = Value::array(); + for (const auto& e : elements) { + if (!e) throw std::runtime_error("Array element is null"); + result.push_back(e->evaluate(context)); + } + return result; + } +}; + +class DictExpr : public Expression { + std::vector, std::shared_ptr>> elements; +public: + DictExpr(const Location & location, std::vector, std::shared_ptr>> && e) + : Expression(location), elements(std::move(e)) {} + Value do_evaluate(const std::shared_ptr & context) const override { + auto result = Value::object(); + for (const auto& [key, value] : elements) { + if (!key) throw std::runtime_error("Dict key is null"); + if (!value) throw std::runtime_error("Dict value is null"); + result.set(key->evaluate(context), value->evaluate(context)); + } + return result; + } +}; + +class SliceExpr : public Expression { +public: + std::shared_ptr start, end; + SliceExpr(const Location & location, std::shared_ptr && s, std::shared_ptr && e) + : Expression(location), start(std::move(s)), end(std::move(e)) {} + Value do_evaluate(const std::shared_ptr &) const override { + throw std::runtime_error("SliceExpr not implemented"); + } +}; + +class SubscriptExpr : public Expression { + std::shared_ptr base; + std::shared_ptr index; +public: + SubscriptExpr(const Location & location, std::shared_ptr && b, std::shared_ptr && i) + : Expression(location), base(std::move(b)), index(std::move(i)) {} + Value do_evaluate(const std::shared_ptr & context) const override { + if (!base) throw std::runtime_error("SubscriptExpr.base is null"); + if (!index) throw std::runtime_error("SubscriptExpr.index is null"); + auto target_value = base->evaluate(context); + if (auto slice = dynamic_cast(index.get())) { + auto start = slice->start ? slice->start->evaluate(context).get() : 0; + auto end = slice->end ? slice->end->evaluate(context).get() : (int64_t) target_value.size(); + if (target_value.is_string()) { + std::string s = target_value.get(); + if (start < 0) start = s.size() + start; + if (end < 0) end = s.size() + end; + return s.substr(start, end - start); + } else if (target_value.is_array()) { + if (start < 0) start = target_value.size() + start; + if (end < 0) end = target_value.size() + end; + auto result = Value::array(); + for (auto i = start; i < end; ++i) { + result.push_back(target_value.at(i)); + } + return result; + } else { + throw std::runtime_error(target_value.is_null() ? "Cannot subscript null" : "Subscripting only supported on arrays and strings"); + } + } else { + auto index_value = index->evaluate(context); + if (target_value.is_null()) { + if (auto t = dynamic_cast(base.get())) { + throw std::runtime_error("'" + t->get_name() + "' is " + (context->contains(t->get_name()) ? "null" : "not defined")); + } + throw std::runtime_error("Trying to access property '" + index_value.dump() + "' on null!"); + } + return target_value.get(index_value); + } + } +}; + +class UnaryOpExpr : public Expression { +public: + enum class Op { Plus, Minus, LogicalNot, Expansion, ExpansionDict }; + std::shared_ptr expr; + Op op; + UnaryOpExpr(const Location & location, std::shared_ptr && e, Op o) + : Expression(location), expr(std::move(e)), op(o) {} + Value do_evaluate(const std::shared_ptr & context) const override { + if (!expr) throw std::runtime_error("UnaryOpExpr.expr is null"); + auto e = expr->evaluate(context); + switch (op) { + case Op::Plus: return e; + case Op::Minus: return -e; + case Op::LogicalNot: return !e.to_bool(); + case Op::Expansion: + case Op::ExpansionDict: + throw std::runtime_error("Expansion operator is only supported in function calls and collections"); + + } + throw std::runtime_error("Unknown unary operator"); + } +}; + +class BinaryOpExpr : public Expression { +public: + enum class Op { StrConcat, Add, Sub, Mul, MulMul, Div, DivDiv, Mod, Eq, Ne, Lt, Gt, Le, Ge, And, Or, In, NotIn, Is, IsNot }; +private: + std::shared_ptr left; + std::shared_ptr right; + Op op; +public: + BinaryOpExpr(const Location & location, std::shared_ptr && l, std::shared_ptr && r, Op o) + : Expression(location), left(std::move(l)), right(std::move(r)), op(o) {} + Value do_evaluate(const std::shared_ptr & context) const override { + if (!left) throw std::runtime_error("BinaryOpExpr.left is null"); + if (!right) throw std::runtime_error("BinaryOpExpr.right is null"); + auto l = left->evaluate(context); + + auto do_eval = [&](const Value & l) -> Value { + if (op == Op::Is || op == Op::IsNot) { + auto t = dynamic_cast(right.get()); + if (!t) throw std::runtime_error("Right side of 'is' operator must be a variable"); + + auto eval = [&]() { + const auto & name = t->get_name(); + if (name == "none") return l.is_null(); + if (name == "boolean") return l.is_boolean(); + if (name == "integer") return l.is_number_integer(); + if (name == "float") return l.is_number_float(); + if (name == "number") return l.is_number(); + if (name == "string") return l.is_string(); + if (name == "mapping") return l.is_object(); + if (name == "iterable") return l.is_iterable(); + if (name == "sequence") return l.is_array(); + if (name == "defined") return !l.is_null(); + throw std::runtime_error("Unknown type for 'is' operator: " + name); + }; + auto value = eval(); + return Value(op == Op::Is ? value : !value); + } + + if (op == Op::And) { + if (!l.to_bool()) return Value(false); + return right->evaluate(context).to_bool(); + } else if (op == Op::Or) { + if (l.to_bool()) return l; + return right->evaluate(context); + } + + auto r = right->evaluate(context); + switch (op) { + case Op::StrConcat: return l.to_str() + r.to_str(); + case Op::Add: return l + r; + case Op::Sub: return l - r; + case Op::Mul: return l * r; + case Op::Div: return l / r; + case Op::MulMul: return std::pow(l.get(), r.get()); + case Op::DivDiv: return l.get() / r.get(); + case Op::Mod: return l.get() % r.get(); + case Op::Eq: return l == r; + case Op::Ne: return l != r; + case Op::Lt: return l < r; + case Op::Gt: return l > r; + case Op::Le: return l <= r; + case Op::Ge: return l >= r; + case Op::In: return (r.is_array() || r.is_object()) && r.contains(l); + case Op::NotIn: return !(r.is_array() && r.contains(l)); + default: break; + } + throw std::runtime_error("Unknown binary operator"); + }; + + if (l.is_callable()) { + return Value::callable([l, do_eval](const std::shared_ptr & context, ArgumentsValue & args) { + auto ll = l.call(context, args); + return do_eval(ll); //args[0].second); + }); + } else { + return do_eval(l); + } + } +}; + +struct ArgumentsExpression { + std::vector> args; + std::vector>> kwargs; + + ArgumentsValue evaluate(const std::shared_ptr & context) const { + ArgumentsValue vargs; + for (const auto& arg : this->args) { + if (auto un_expr = std::dynamic_pointer_cast(arg)) { + if (un_expr->op == UnaryOpExpr::Op::Expansion) { + auto array = un_expr->expr->evaluate(context); + if (!array.is_array()) { + throw std::runtime_error("Expansion operator only supported on arrays"); + } + array.for_each([&](Value & value) { + vargs.args.push_back(value); + }); + continue; + } else if (un_expr->op == UnaryOpExpr::Op::ExpansionDict) { + auto dict = un_expr->expr->evaluate(context); + if (!dict.is_object()) { + throw std::runtime_error("ExpansionDict operator only supported on objects"); + } + dict.for_each([&](const Value & key) { + vargs.kwargs.push_back({key.get(), dict.at(key)}); + }); + continue; + } + } + vargs.args.push_back(arg->evaluate(context)); + } + for (const auto& [name, value] : this->kwargs) { + vargs.kwargs.push_back({name, value->evaluate(context)}); + } + return vargs; + } +}; + +static std::string strip(const std::string & s) { + auto start = s.find_first_not_of(" \t\n\r"); + if (start == std::string::npos) return ""; + auto end = s.find_last_not_of(" \t\n\r"); + return s.substr(start, end - start + 1); +} + +static std::string html_escape(const std::string & s) { + std::string result; + result.reserve(s.size()); + for (const auto & c : s) { + switch (c) { + case '&': result += "&"; break; + case '<': result += "<"; break; + case '>': result += ">"; break; + case '"': result += """; break; + case '\'': result += "'"; break; + default: result += c; break; + } + } + return result; +} + +class MethodCallExpr : public Expression { + std::shared_ptr object; + std::shared_ptr method; + ArgumentsExpression args; +public: + MethodCallExpr(const Location & location, std::shared_ptr && obj, std::shared_ptr && m, ArgumentsExpression && a) + : Expression(location), object(std::move(obj)), method(std::move(m)), args(std::move(a)) {} + Value do_evaluate(const std::shared_ptr & context) const override { + if (!object) throw std::runtime_error("MethodCallExpr.object is null"); + if (!method) throw std::runtime_error("MethodCallExpr.method is null"); + auto obj = object->evaluate(context); + auto vargs = args.evaluate(context); + if (obj.is_null()) { + throw std::runtime_error("Trying to call method '" + method->get_name() + "' on null"); + } + if (obj.is_array()) { + if (method->get_name() == "append") { + vargs.expectArgs("append method", {1, 1}, {0, 0}); + obj.push_back(vargs.args[0]); + return Value(); + } else if (method->get_name() == "pop") { + vargs.expectArgs("pop method", {0, 1}, {0, 0}); + return obj.pop(vargs.args.empty() ? Value() : vargs.args[0]); + } else if (method->get_name() == "insert") { + vargs.expectArgs("insert method", {2, 2}, {0, 0}); + auto index = vargs.args[0].get(); + if (index < 0 || index > (int64_t) obj.size()) throw std::runtime_error("Index out of range for insert method"); + obj.insert(index, vargs.args[1]); + return Value(); + } + } else if (obj.is_object()) { + if (method->get_name() == "items") { + vargs.expectArgs("items method", {0, 0}, {0, 0}); + auto result = Value::array(); + for (const auto& key : obj.keys()) { + result.push_back(Value::array({key, obj.at(key)})); + } + return result; + } else if (method->get_name() == "pop") { + vargs.expectArgs("pop method", {1, 1}, {0, 0}); + return obj.pop(vargs.args[0]); + } else if (method->get_name() == "get") { + vargs.expectArgs("get method", {1, 2}, {0, 0}); + auto key = vargs.args[0]; + if (vargs.args.size() == 1) { + return obj.contains(key) ? obj.at(key) : Value(); + } else { + return obj.contains(key) ? obj.at(key) : vargs.args[1]; + } + } else if (obj.contains(method->get_name())) { + auto callable = obj.at(method->get_name()); + if (!callable.is_callable()) { + throw std::runtime_error("Property '" + method->get_name() + "' is not callable"); + } + return callable.call(context, vargs); + } + } else if (obj.is_string()) { + auto str = obj.get(); + if (method->get_name() == "strip") { + vargs.expectArgs("strip method", {0, 0}, {0, 0}); + return Value(strip(str)); + } else if (method->get_name() == "endswith") { + vargs.expectArgs("endswith method", {1, 1}, {0, 0}); + auto suffix = vargs.args[0].get(); + return suffix.length() <= str.length() && std::equal(suffix.rbegin(), suffix.rend(), str.rbegin()); + } else if (method->get_name() == "title") { + vargs.expectArgs("title method", {0, 0}, {0, 0}); + auto res = str; + for (size_t i = 0, n = res.size(); i < n; ++i) { + if (i == 0 || std::isspace(res[i - 1])) res[i] = std::toupper(res[i]); + else res[i] = std::tolower(res[i]); + } + return res; + } + } + throw std::runtime_error("Unknown method: " + method->get_name()); + } +}; + +class CallExpr : public Expression { +public: + std::shared_ptr object; + ArgumentsExpression args; + CallExpr(const Location & location, std::shared_ptr && obj, ArgumentsExpression && a) + : Expression(location), object(std::move(obj)), args(std::move(a)) {} + Value do_evaluate(const std::shared_ptr & context) const override { + if (!object) throw std::runtime_error("CallExpr.object is null"); + auto obj = object->evaluate(context); + if (!obj.is_callable()) { + throw std::runtime_error("Object is not callable: " + obj.dump(2)); + } + auto vargs = args.evaluate(context); + return obj.call(context, vargs); + } +}; + +class FilterExpr : public Expression { + std::vector> parts; +public: + FilterExpr(const Location & location, std::vector> && p) + : Expression(location), parts(std::move(p)) {} + Value do_evaluate(const std::shared_ptr & context) const override { + Value result; + bool first = true; + for (const auto& part : parts) { + if (!part) throw std::runtime_error("FilterExpr.part is null"); + if (first) { + first = false; + result = part->evaluate(context); + } else { + if (auto ce = dynamic_cast(part.get())) { + auto target = ce->object->evaluate(context); + ArgumentsValue args = ce->args.evaluate(context); + args.args.insert(args.args.begin(), result); + result = target.call(context, args); + } else { + auto callable = part->evaluate(context); + ArgumentsValue args; + args.args.insert(args.args.begin(), result); + result = callable.call(context, args); + } + } + } + return result; + } + + void prepend(std::shared_ptr && e) { + parts.insert(parts.begin(), std::move(e)); + } +}; + +class Parser { +private: + using CharIterator = std::string::const_iterator; + + std::shared_ptr template_str; + CharIterator start, end, it; + Options options; + + Parser(const std::shared_ptr& template_str, const Options & options) : template_str(template_str), options(options) { + if (!template_str) throw std::runtime_error("Template string is null"); + start = it = this->template_str->begin(); + end = this->template_str->end(); + } + + bool consumeSpaces(SpaceHandling space_handling = SpaceHandling::Strip) { + if (space_handling == SpaceHandling::Strip) { + while (it != end && std::isspace(*it)) ++it; + } + return true; + } + + std::unique_ptr parseString() { + auto doParse = [&](char quote) -> std::unique_ptr { + if (it == end || *it != quote) return nullptr; + std::string result; + bool escape = false; + for (++it; it != end; ++it) { + if (escape) { + escape = false; + switch (*it) { + case 'n': result += '\n'; break; + case 'r': result += '\r'; break; + case 't': result += '\t'; break; + case 'b': result += '\b'; break; + case 'f': result += '\f'; break; + case '\\': result += '\\'; break; + default: + if (*it == quote) { + result += quote; + } else { + result += *it; + } + break; + } + } else if (*it == '\\') { + escape = true; + } else if (*it == quote) { + ++it; + return std::make_unique(std::move(result)); + } else { + result += *it; + } + } + return nullptr; + }; + + consumeSpaces(); + if (it == end) return nullptr; + if (*it == '"') return doParse('"'); + if (*it == '\'') return doParse('\''); + return nullptr; + } + + json parseNumber(CharIterator& it, const CharIterator& end) { + auto before = it; + consumeSpaces(); + auto start = it; + bool hasDecimal = false; + bool hasExponent = false; + + if (it != end && (*it == '-' || *it == '+')) ++it; + + while (it != end) { + if (std::isdigit(*it)) { + ++it; + } else if (*it == '.') { + if (hasDecimal) throw std::runtime_error("Multiple decimal points"); + hasDecimal = true; + ++it; + } else if (it != start && (*it == 'e' || *it == 'E')) { + if (hasExponent) throw std::runtime_error("Multiple exponents"); + hasExponent = true; + ++it; + } else { + break; + } + } + if (start == it) { + it = before; + return json(); // No valid characters found + } + + std::string str(start, it); + try { + return json::parse(str); + } catch (json::parse_error& e) { + throw std::runtime_error("Failed to parse number: '" + str + "' (" + std::string(e.what()) + ")"); + return json(); + } + } + + /** integer, float, bool, string */ + std::shared_ptr parseConstant() { + auto start = it; + consumeSpaces(); + if (it == end) return nullptr; + if (*it == '"' || *it == '\'') { + auto str = parseString(); + if (str) return std::make_shared(*str); + } + static std::regex prim_tok(R"(true\b|True\b|false\b|False\b|None\b)"); + auto token = consumeToken(prim_tok); + if (!token.empty()) { + if (token == "true" || token == "True") return std::make_shared(true); + if (token == "false" || token == "False") return std::make_shared(false); + if (token == "None") return std::make_shared(nullptr); + throw std::runtime_error("Unknown constant token: " + token); + } + + auto number = parseNumber(it, end); + if (!number.is_null()) return std::make_shared(number); + + it = start; + return nullptr; + } + + class expression_parsing_error : public std::runtime_error { + const CharIterator it; + public: + expression_parsing_error(const std::string & message, const CharIterator & it) + : std::runtime_error(message), it(it) {} + size_t get_pos(const CharIterator & begin) const { + return std::distance(begin, it); + } + }; + + bool peekSymbols(const std::vector & symbols) const { + for (const auto & symbol : symbols) { + if (std::distance(it, end) >= (int64_t) symbol.size() && std::string(it, it + symbol.size()) == symbol) { + return true; + } + } + return false; + } + + std::vector consumeTokenGroups(const std::regex & regex, SpaceHandling space_handling = SpaceHandling::Strip) { + auto start = it; + consumeSpaces(space_handling); + std::smatch match; + if (std::regex_search(it, end, match, regex) && match.position() == 0) { + it += match[0].length(); + std::vector ret; + for (size_t i = 0, n = match.size(); i < n; ++i) { + ret.push_back(match[i].str()); + } + return ret; + } + it = start; + return {}; + } + std::string consumeToken(const std::regex & regex, SpaceHandling space_handling = SpaceHandling::Strip) { + auto start = it; + consumeSpaces(space_handling); + std::smatch match; + if (std::regex_search(it, end, match, regex) && match.position() == 0) { + it += match[0].length(); + return match[0].str(); + } + it = start; + return ""; + } + + std::string consumeToken(const std::string & token, SpaceHandling space_handling = SpaceHandling::Strip) { + auto start = it; + consumeSpaces(space_handling); + if (std::distance(it, end) >= (int64_t) token.size() && std::string(it, it + token.size()) == token) { + it += token.size(); + return token; + } + it = start; + return ""; + } + + std::shared_ptr parseExpression(bool allow_if_expr = true) { + auto left = parseLogicalOr(); + if (it == end) return left; + + if (!allow_if_expr) return left; + + static std::regex if_tok(R"(if\b)"); + if (consumeToken(if_tok).empty()) { + return left; + } + + auto location = get_location(); + auto [condition, else_expr] = parseIfExpression(); + return std::make_shared(location, std::move(condition), std::move(left), std::move(else_expr)); + } + + Location get_location() const { + return {template_str, (size_t) std::distance(start, it)}; + } + + std::pair, std::shared_ptr> parseIfExpression() { + auto condition = parseLogicalOr(); + if (!condition) throw std::runtime_error("Expected condition expression"); + + static std::regex else_tok(R"(else\b)"); + std::shared_ptr else_expr; + if (!consumeToken(else_tok).empty()) { + else_expr = parseExpression(); + if (!else_expr) throw std::runtime_error("Expected 'else' expression"); + } + return std::pair(std::move(condition), std::move(else_expr)); + } + + std::shared_ptr parseLogicalOr() { + auto left = parseLogicalAnd(); + if (!left) throw std::runtime_error("Expected left side of 'logical or' expression"); + + static std::regex or_tok(R"(or\b)"); + auto location = get_location(); + while (!consumeToken(or_tok).empty()) { + auto right = parseLogicalAnd(); + if (!right) throw std::runtime_error("Expected right side of 'or' expression"); + left = std::make_shared(location, std::move(left), std::move(right), BinaryOpExpr::Op::Or); + } + return left; + } + + std::shared_ptr parseLogicalNot() { + static std::regex not_tok(R"(not\b)"); + auto location = get_location(); + + if (!consumeToken(not_tok).empty()) { + auto sub = parseLogicalNot(); + if (!sub) throw std::runtime_error("Expected expression after 'not' keyword"); + return std::make_shared(location, std::move(sub), UnaryOpExpr::Op::LogicalNot); + } + return parseLogicalCompare(); + } + + std::shared_ptr parseLogicalAnd() { + auto left = parseLogicalNot(); + if (!left) throw std::runtime_error("Expected left side of 'logical and' expression"); + + static std::regex and_tok(R"(and\b)"); + auto location = get_location(); + while (!consumeToken(and_tok).empty()) { + auto right = parseLogicalNot(); + if (!right) throw std::runtime_error("Expected right side of 'and' expression"); + left = std::make_shared(location, std::move(left), std::move(right), BinaryOpExpr::Op::And); + } + return left; + } + + std::shared_ptr parseLogicalCompare() { + auto left = parseStringConcat(); + if (!left) throw std::runtime_error("Expected left side of 'logical compare' expression"); + + static std::regex compare_tok(R"(==|!=|<=?|>=?|in\b|is\b|not[\r\n\s]+in\b)"); + static std::regex not_tok(R"(not\b)"); + std::string op_str; + while (!(op_str = consumeToken(compare_tok)).empty()) { + auto location = get_location(); + if (op_str == "is") { + auto negated = !consumeToken(not_tok).empty(); + + auto identifier = parseIdentifier(); + if (!identifier) throw std::runtime_error("Expected identifier after 'is' keyword"); + + return std::make_shared( + left->location, + std::move(left), std::move(identifier), + negated ? BinaryOpExpr::Op::IsNot : BinaryOpExpr::Op::Is); + } + auto right = parseStringConcat(); + if (!right) throw std::runtime_error("Expected right side of 'logical compare' expression"); + BinaryOpExpr::Op op; + if (op_str == "==") op = BinaryOpExpr::Op::Eq; + else if (op_str == "!=") op = BinaryOpExpr::Op::Ne; + else if (op_str == "<") op = BinaryOpExpr::Op::Lt; + else if (op_str == ">") op = BinaryOpExpr::Op::Gt; + else if (op_str == "<=") op = BinaryOpExpr::Op::Le; + else if (op_str == ">=") op = BinaryOpExpr::Op::Ge; + else if (op_str == "in") op = BinaryOpExpr::Op::In; + else if (op_str.substr(0, 3) == "not") op = BinaryOpExpr::Op::NotIn; + else throw std::runtime_error("Unknown comparison operator: " + op_str); + left = std::make_shared(get_location(), std::move(left), std::move(right), op); + } + return left; + } + + Expression::Parameters parseParameters() { + consumeSpaces(); + if (consumeToken("(").empty()) throw std::runtime_error("Expected opening parenthesis in param list"); + + Expression::Parameters result; + + while (it != end) { + if (!consumeToken(")").empty()) { + return result; + } + auto expr = parseExpression(); + if (!expr) throw std::runtime_error("Expected expression in call args"); + + if (auto ident = dynamic_cast(expr.get())) { + if (!consumeToken("=").empty()) { + auto value = parseExpression(); + if (!value) throw std::runtime_error("Expected expression in for named arg"); + result.emplace_back(ident->get_name(), std::move(value)); + } else { + result.emplace_back(ident->get_name(), nullptr); + } + } else { + result.emplace_back(std::string(), std::move(expr)); + } + if (consumeToken(",").empty()) { + if (consumeToken(")").empty()) { + throw std::runtime_error("Expected closing parenthesis in call args"); + } + return result; + } + } + throw std::runtime_error("Expected closing parenthesis in call args"); + } + + ArgumentsExpression parseCallArgs() { + consumeSpaces(); + if (consumeToken("(").empty()) throw std::runtime_error("Expected opening parenthesis in call args"); + + ArgumentsExpression result; + + while (it != end) { + if (!consumeToken(")").empty()) { + return result; + } + auto expr = parseExpression(); + if (!expr) throw std::runtime_error("Expected expression in call args"); + + if (auto ident = dynamic_cast(expr.get())) { + if (!consumeToken("=").empty()) { + auto value = parseExpression(); + if (!value) throw std::runtime_error("Expected expression in for named arg"); + result.kwargs.emplace_back(ident->get_name(), std::move(value)); + } else { + result.args.emplace_back(std::move(expr)); + } + } else { + result.args.emplace_back(std::move(expr)); + } + if (consumeToken(",").empty()) { + if (consumeToken(")").empty()) { + throw std::runtime_error("Expected closing parenthesis in call args"); + } + return result; + } + } + throw std::runtime_error("Expected closing parenthesis in call args"); + } + + std::shared_ptr parseIdentifier() { + static std::regex ident_regex(R"((?!(?:not|is|and|or|del)\b)[a-zA-Z_]\w*)"); + auto location = get_location(); + auto ident = consumeToken(ident_regex); + if (ident.empty()) + return nullptr; + return std::make_shared(location, ident); + } + + std::shared_ptr parseStringConcat() { + auto left = parseMathPow(); + if (!left) throw std::runtime_error("Expected left side of 'string concat' expression"); + + static std::regex concat_tok(R"(~(?!\}))"); + if (!consumeToken(concat_tok).empty()) { + auto right = parseLogicalAnd(); + if (!right) throw std::runtime_error("Expected right side of 'string concat' expression"); + left = std::make_shared(get_location(), std::move(left), std::move(right), BinaryOpExpr::Op::StrConcat); + } + return left; + } + + std::shared_ptr parseMathPow() { + auto left = parseMathPlusMinus(); + if (!left) throw std::runtime_error("Expected left side of 'math pow' expression"); + + while (!consumeToken("**").empty()) { + auto right = parseMathPlusMinus(); + if (!right) throw std::runtime_error("Expected right side of 'math pow' expression"); + left = std::make_shared(get_location(), std::move(left), std::move(right), BinaryOpExpr::Op::MulMul); + } + return left; + } + + std::shared_ptr parseMathPlusMinus() { + static std::regex plus_minus_tok(R"(\+|-(?![}%#]\}))"); + + auto left = parseMathMulDiv(); + if (!left) throw std::runtime_error("Expected left side of 'math plus/minus' expression"); + std::string op_str; + while (!(op_str = consumeToken(plus_minus_tok)).empty()) { + auto right = parseMathMulDiv(); + if (!right) throw std::runtime_error("Expected right side of 'math plus/minus' expression"); + auto op = op_str == "+" ? BinaryOpExpr::Op::Add : BinaryOpExpr::Op::Sub; + left = std::make_shared(get_location(), std::move(left), std::move(right), op); + } + return left; + } + + std::shared_ptr parseMathMulDiv() { + auto left = parseMathUnaryPlusMinus(); + if (!left) throw std::runtime_error("Expected left side of 'math mul/div' expression"); + + static std::regex mul_div_tok(R"(\*\*?|//?|%(?!\}))"); + std::string op_str; + while (!(op_str = consumeToken(mul_div_tok)).empty()) { + auto right = parseMathUnaryPlusMinus(); + if (!right) throw std::runtime_error("Expected right side of 'math mul/div' expression"); + auto op = op_str == "*" ? BinaryOpExpr::Op::Mul + : op_str == "**" ? BinaryOpExpr::Op::MulMul + : op_str == "/" ? BinaryOpExpr::Op::Div + : op_str == "//" ? BinaryOpExpr::Op::DivDiv + : BinaryOpExpr::Op::Mod; + left = std::make_shared(get_location(), std::move(left), std::move(right), op); + } + + if (!consumeToken("|").empty()) { + auto expr = parseMathMulDiv(); + if (auto filter = dynamic_cast(expr.get())) { + filter->prepend(std::move(left)); + return expr; + } else { + std::vector> parts; + parts.emplace_back(std::move(left)); + parts.emplace_back(std::move(expr)); + return std::make_shared(get_location(), std::move(parts)); + } + } + return left; + } + + std::shared_ptr call_func(const std::string & name, ArgumentsExpression && args) const { + return std::make_shared(get_location(), std::make_shared(get_location(), name), std::move(args)); + } + + std::shared_ptr parseMathUnaryPlusMinus() { + static std::regex unary_plus_minus_tok(R"(\+|-(?![}%#]\}))"); + auto op_str = consumeToken(unary_plus_minus_tok); + auto expr = parseExpansion(); + if (!expr) throw std::runtime_error("Expected expr of 'unary plus/minus/expansion' expression"); + + if (!op_str.empty()) { + auto op = op_str == "+" ? UnaryOpExpr::Op::Plus : UnaryOpExpr::Op::Minus; + return std::make_shared(get_location(), std::move(expr), op); + } + return expr; + } + + std::shared_ptr parseExpansion() { + static std::regex expansion_tok(R"(\*\*?)"); + auto op_str = consumeToken(expansion_tok); + auto expr = parseValueExpression(); + if (op_str.empty()) return expr; + if (!expr) throw std::runtime_error("Expected expr of 'expansion' expression"); + return std::make_shared(get_location(), std::move(expr), op_str == "*" ? UnaryOpExpr::Op::Expansion : UnaryOpExpr::Op::ExpansionDict); + } + + std::shared_ptr parseValueExpression() { + auto parseValue = [&]() -> std::shared_ptr { + auto location = get_location(); + auto constant = parseConstant(); + if (constant) return std::make_shared(location, *constant); + + static std::regex null_regex(R"(null\b)"); + if (!consumeToken(null_regex).empty()) return std::make_shared(location, Value()); + + auto identifier = parseIdentifier(); + if (identifier) return identifier; + + auto braced = parseBracedExpressionOrArray(); + if (braced) return braced; + + auto array = parseArray(); + if (array) return array; + + auto dictionary = parseDictionary(); + if (dictionary) return dictionary; + + throw std::runtime_error("Expected value expression"); + }; + + auto value = parseValue(); + + while (it != end && consumeSpaces() && peekSymbols({ "[", "." })) { + if (!consumeToken("[").empty()) { + std::shared_ptr index; + if (!consumeToken(":").empty()) { + auto slice_end = parseExpression(); + index = std::make_shared(slice_end->location, nullptr, std::move(slice_end)); + } else { + auto slice_start = parseExpression(); + if (!consumeToken(":").empty()) { + consumeSpaces(); + if (peekSymbols({ "]" })) { + index = std::make_shared(slice_start->location, std::move(slice_start), nullptr); + } else { + auto slice_end = parseExpression(); + index = std::make_shared(slice_start->location, std::move(slice_start), std::move(slice_end)); + } + } else { + index = std::move(slice_start); + } + } + if (!index) throw std::runtime_error("Empty index in subscript"); + if (consumeToken("]").empty()) throw std::runtime_error("Expected closing bracket in subscript"); + + value = std::make_shared(value->location, std::move(value), std::move(index)); + } else if (!consumeToken(".").empty()) { + auto identifier = parseIdentifier(); + if (!identifier) throw std::runtime_error("Expected identifier in subscript"); + + consumeSpaces(); + if (peekSymbols({ "(" })) { + auto callParams = parseCallArgs(); + value = std::make_shared(identifier->location, std::move(value), std::move(identifier), std::move(callParams)); + } else { + auto key = std::make_shared(identifier->location, Value(identifier->get_name())); + value = std::make_shared(identifier->location, std::move(value), std::move(key)); + } + } + consumeSpaces(); + } + + if (peekSymbols({ "(" })) { + auto location = get_location(); + auto callParams = parseCallArgs(); + value = std::make_shared(location, std::move(value), std::move(callParams)); + } + return value; + } + + std::shared_ptr parseBracedExpressionOrArray() { + if (consumeToken("(").empty()) return nullptr; + + auto expr = parseExpression(); + if (!expr) throw std::runtime_error("Expected expression in braced expression"); + + if (!consumeToken(")").empty()) { + return expr; // Drop the parentheses + } + + std::vector> tuple; + tuple.emplace_back(std::move(expr)); + + while (it != end) { + if (consumeToken(",").empty()) throw std::runtime_error("Expected comma in tuple"); + auto next = parseExpression(); + if (!next) throw std::runtime_error("Expected expression in tuple"); + tuple.push_back(std::move(next)); + + if (!consumeToken(")").empty()) { + return std::make_shared(get_location(), std::move(tuple)); + } + } + throw std::runtime_error("Expected closing parenthesis"); + } + + std::shared_ptr parseArray() { + if (consumeToken("[").empty()) return nullptr; + + std::vector> elements; + if (!consumeToken("]").empty()) { + return std::make_shared(get_location(), std::move(elements)); + } + auto first_expr = parseExpression(); + if (!first_expr) throw std::runtime_error("Expected first expression in array"); + elements.push_back(std::move(first_expr)); + + while (it != end) { + if (!consumeToken(",").empty()) { + auto expr = parseExpression(); + if (!expr) throw std::runtime_error("Expected expression in array"); + elements.push_back(std::move(expr)); + } else if (!consumeToken("]").empty()) { + return std::make_shared(get_location(), std::move(elements)); + } else { + throw std::runtime_error("Expected comma or closing bracket in array"); + } + } + throw std::runtime_error("Expected closing bracket"); + } + + std::shared_ptr parseDictionary() { + if (consumeToken("{").empty()) return nullptr; + + std::vector, std::shared_ptr>> elements; + if (!consumeToken("}").empty()) { + return std::make_shared(get_location(), std::move(elements)); + } + + auto parseKeyValuePair = [&]() { + auto key = parseExpression(); + if (!key) throw std::runtime_error("Expected key in dictionary"); + if (consumeToken(":").empty()) throw std::runtime_error("Expected colon betweek key & value in dictionary"); + auto value = parseExpression(); + if (!value) throw std::runtime_error("Expected value in dictionary"); + elements.emplace_back(std::pair(std::move(key), std::move(value))); + }; + + parseKeyValuePair(); + + while (it != end) { + if (!consumeToken(",").empty()) { + parseKeyValuePair(); + } else if (!consumeToken("}").empty()) { + return std::make_shared(get_location(), std::move(elements)); + } else { + throw std::runtime_error("Expected comma or closing brace in dictionary"); + } + } + throw std::runtime_error("Expected closing brace"); + } + + SpaceHandling parsePreSpace(const std::string& s) const { + if (s == "-") + return SpaceHandling::Strip; + return SpaceHandling::Keep; + } + + SpaceHandling parsePostSpace(const std::string& s) const { + if (s == "-") return SpaceHandling::Strip; + return SpaceHandling::Keep; + } + + using TemplateTokenVector = std::vector>; + using TemplateTokenIterator = TemplateTokenVector::const_iterator; + + std::vector parseVarNames() { + static std::regex varnames_regex(R"(((?:\w+)(?:[\r\n\s]*,[\r\n\s]*(?:\w+))*)[\r\n\s]*)"); + + std::vector group; + if ((group = consumeTokenGroups(varnames_regex)).empty()) throw std::runtime_error("Expected variable names"); + std::vector varnames; + std::istringstream iss(group[1]); + std::string varname; + while (std::getline(iss, varname, ',')) { + varnames.push_back(strip(varname)); + } + return varnames; + } + + std::runtime_error unexpected(const TemplateToken & token) const { + return std::runtime_error("Unexpected " + TemplateToken::typeToString(token.type) + + error_location_suffix(*template_str, token.location.pos)); + } + std::runtime_error unterminated(const TemplateToken & token) const { + return std::runtime_error("Unterminated " + TemplateToken::typeToString(token.type) + + error_location_suffix(*template_str, token.location.pos)); + } + + TemplateTokenVector tokenize() { + static std::regex comment_tok(R"(\{#([-~]?)(.*?)([-~]?)#\})"); + static std::regex expr_open_regex(R"(\{\{([-~])?)"); + static std::regex block_open_regex(R"(^\{%([-~])?[\s\n\r]*)"); + static std::regex block_keyword_tok(R"((if|else|elif|endif|for|endfor|generation|endgeneration|set|endset|block|endblock|macro|endmacro|filter|endfilter)\b)"); + static std::regex non_text_open_regex(R"(\{\{|\{%|\{#)"); + static std::regex expr_close_regex(R"([\s\n\r]*([-~])?\}\})"); + static std::regex block_close_regex(R"([\s\n\r]*([-~])?%\})"); + + TemplateTokenVector tokens; + std::vector group; + std::string text; + std::smatch match; + + try { + while (it != end) { + auto location = get_location(); + + if (!(group = consumeTokenGroups(comment_tok, SpaceHandling::Keep)).empty()) { + auto pre_space = parsePreSpace(group[1]); + auto content = group[2]; + auto post_space = parsePostSpace(group[3]); + tokens.push_back(std::make_unique(location, pre_space, post_space, content)); + } else if (!(group = consumeTokenGroups(expr_open_regex, SpaceHandling::Keep)).empty()) { + auto pre_space = parsePreSpace(group[1]); + auto expr = parseExpression(); + + if ((group = consumeTokenGroups(expr_close_regex)).empty()) { + throw std::runtime_error("Expected closing expression tag"); + } + + auto post_space = parsePostSpace(group[1]); + tokens.push_back(std::make_unique(location, pre_space, post_space, std::move(expr))); + } else if (!(group = consumeTokenGroups(block_open_regex, SpaceHandling::Keep)).empty()) { + auto pre_space = parsePreSpace(group[1]); + + std::string keyword; + + auto parseBlockClose = [&]() -> SpaceHandling { + if ((group = consumeTokenGroups(block_close_regex)).empty()) throw std::runtime_error("Expected closing block tag"); + return parsePostSpace(group[1]); + }; + + if ((keyword = consumeToken(block_keyword_tok)).empty()) throw std::runtime_error("Expected block keyword"); + + if (keyword == "if") { + auto condition = parseExpression(); + if (!condition) throw std::runtime_error("Expected condition in if block"); + + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space, std::move(condition))); + } else if (keyword == "elif") { + auto condition = parseExpression(); + if (!condition) throw std::runtime_error("Expected condition in elif block"); + + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space, std::move(condition))); + } else if (keyword == "else") { + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space)); + } else if (keyword == "endif") { + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space)); + } else if (keyword == "for") { + static std::regex recursive_tok(R"(recursive\b)"); + static std::regex if_tok(R"(if\b)"); + + auto varnames = parseVarNames(); + static std::regex in_tok(R"(in\b)"); + if (consumeToken(in_tok).empty()) throw std::runtime_error("Expected 'in' keyword in for block"); + auto iterable = parseExpression(/* allow_if_expr = */ false); + if (!iterable) throw std::runtime_error("Expected iterable in for block"); + + std::shared_ptr condition; + if (!consumeToken(if_tok).empty()) { + condition = parseExpression(); + } + auto recursive = !consumeToken(recursive_tok).empty(); + + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space, std::move(varnames), std::move(iterable), std::move(condition), recursive)); + } else if (keyword == "endfor") { + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space)); + } else if (keyword == "generation") { + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space)); + } else if (keyword == "endgeneration") { + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space)); + } else if (keyword == "set") { + static std::regex namespaced_var_regex(R"((\w+)[\s\n\r]*\.[\s\n\r]*(\w+))"); + + std::string ns; + std::vector var_names; + std::shared_ptr value; + if (!(group = consumeTokenGroups(namespaced_var_regex)).empty()) { + ns = group[1]; + var_names.push_back(group[2]); + + if (consumeToken("=").empty()) throw std::runtime_error("Expected equals sign in set block"); + + value = parseExpression(); + if (!value) throw std::runtime_error("Expected value in set block"); + } else { + var_names = parseVarNames(); + + if (!consumeToken("=").empty()) { + value = parseExpression(); + if (!value) throw std::runtime_error("Expected value in set block"); + } + } + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space, ns, var_names, std::move(value))); + } else if (keyword == "endset") { + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space)); + } else if (keyword == "macro") { + auto macroname = parseIdentifier(); + if (!macroname) throw std::runtime_error("Expected macro name in macro block"); + auto params = parseParameters(); + + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space, std::move(macroname), std::move(params))); + } else if (keyword == "endmacro") { + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space)); + } else if (keyword == "filter") { + auto filter = parseExpression(); + if (!filter) throw std::runtime_error("Expected expression in filter block"); + + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space, std::move(filter))); + } else if (keyword == "endfilter") { + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space)); + } else { + throw std::runtime_error("Unexpected block: " + keyword); + } + } else if (std::regex_search(it, end, match, non_text_open_regex)) { + auto text_end = it + match.position(); + text = std::string(it, text_end); + it = text_end; + tokens.push_back(std::make_unique(location, SpaceHandling::Keep, SpaceHandling::Keep, text)); + } else { + text = std::string(it, end); + it = end; + tokens.push_back(std::make_unique(location, SpaceHandling::Keep, SpaceHandling::Keep, text)); + } + } + return tokens; + } catch (const std::exception & e) { + throw std::runtime_error(e.what() + error_location_suffix(*template_str, std::distance(start, it))); + } + } + + std::shared_ptr parseTemplate( + const TemplateTokenIterator & begin, + TemplateTokenIterator & it, + const TemplateTokenIterator & end, + bool fully = false) const { + std::vector> children; + while (it != end) { + const auto start = it; + const auto & token = *(it++); + if (auto if_token = dynamic_cast(token.get())) { + std::vector, std::shared_ptr>> cascade; + cascade.emplace_back(std::move(if_token->condition), parseTemplate(begin, it, end)); + + while (it != end && (*it)->type == TemplateToken::Type::Elif) { + auto elif_token = dynamic_cast((*(it++)).get()); + cascade.emplace_back(std::move(elif_token->condition), parseTemplate(begin, it, end)); + } + + if (it != end && (*it)->type == TemplateToken::Type::Else) { + cascade.emplace_back(nullptr, parseTemplate(begin, ++it, end)); + } + if (it == end || (*(it++))->type != TemplateToken::Type::EndIf) { + throw unterminated(**start); + } + children.emplace_back(std::make_shared(token->location, std::move(cascade))); + } else if (auto for_token = dynamic_cast(token.get())) { + auto body = parseTemplate(begin, it, end); + auto else_body = std::shared_ptr(); + if (it != end && (*it)->type == TemplateToken::Type::Else) { + else_body = parseTemplate(begin, ++it, end); + } + if (it == end || (*(it++))->type != TemplateToken::Type::EndFor) { + throw unterminated(**start); + } + children.emplace_back(std::make_shared(token->location, std::move(for_token->var_names), std::move(for_token->iterable), std::move(for_token->condition), std::move(body), for_token->recursive, std::move(else_body))); + } else if (dynamic_cast(token.get())) { + auto body = parseTemplate(begin, it, end); + if (it == end || (*(it++))->type != TemplateToken::Type::EndGeneration) { + throw unterminated(**start); + } + // Treat as a no-op, as our scope is templates for inference, not training (`{% generation %}` wraps generated tokens for masking). + children.emplace_back(std::move(body)); + } else if (auto text_token = dynamic_cast(token.get())) { + SpaceHandling pre_space = (it - 1) != begin ? (*(it - 2))->post_space : SpaceHandling::Keep; + SpaceHandling post_space = it != end ? (*it)->pre_space : SpaceHandling::Keep; + + auto text = text_token->text; + if (post_space == SpaceHandling::Strip) { + static std::regex trailing_space_regex(R"((\s|\r|\n)+$)"); + text = std::regex_replace(text, trailing_space_regex, ""); + } else if (options.lstrip_blocks && it != end) { + auto i = text.size(); + while (i > 0 && (text[i - 1] == ' ' || text[i - 1] == '\t')) i--; + if ((i == 0 && (it - 1) == begin) || (i > 0 && text[i - 1] == '\n')) { + text.resize(i); + } + } + if (pre_space == SpaceHandling::Strip) { + static std::regex leading_space_regex(R"(^(\s|\r|\n)+)"); + text = std::regex_replace(text, leading_space_regex, ""); + } else if (options.trim_blocks && (it - 1) != begin && !dynamic_cast((*(it - 2)).get())) { + if (text.length() > 0 && text[0] == '\n') { + text.erase(0, 1); + } + } + if (it == end && !options.keep_trailing_newline) { + auto i = text.size(); + if (i > 0 && text[i - 1] == '\n') { + i--; + if (i > 0 && text[i - 1] == '\r') i--; + text.resize(i); + } + } + children.emplace_back(std::make_shared(token->location, text)); + } else if (auto expr_token = dynamic_cast(token.get())) { + children.emplace_back(std::make_shared(token->location, std::move(expr_token->expr))); + } else if (auto set_token = dynamic_cast(token.get())) { + if (set_token->value) { + children.emplace_back(std::make_shared(token->location, set_token->ns, set_token->var_names, std::move(set_token->value))); + } else { + auto value_template = parseTemplate(begin, it, end); + if (it == end || (*(it++))->type != TemplateToken::Type::EndSet) { + throw unterminated(**start); + } + if (!set_token->ns.empty()) throw std::runtime_error("Namespaced set not supported in set with template value"); + if (set_token->var_names.size() != 1) throw std::runtime_error("Structural assignment not supported in set with template value"); + auto & name = set_token->var_names[0]; + children.emplace_back(std::make_shared(token->location, name, std::move(value_template))); + } + } else if (auto macro_token = dynamic_cast(token.get())) { + auto body = parseTemplate(begin, it, end); + if (it == end || (*(it++))->type != TemplateToken::Type::EndMacro) { + throw unterminated(**start); + } + children.emplace_back(std::make_shared(token->location, std::move(macro_token->name), std::move(macro_token->params), std::move(body))); + } else if (auto filter_token = dynamic_cast(token.get())) { + auto body = parseTemplate(begin, it, end); + if (it == end || (*(it++))->type != TemplateToken::Type::EndFilter) { + throw unterminated(**start); + } + children.emplace_back(std::make_shared(token->location, std::move(filter_token->filter), std::move(body))); + } else if (dynamic_cast(token.get())) { + // Ignore comments + } else if (dynamic_cast(token.get()) + || dynamic_cast(token.get()) + || dynamic_cast(token.get()) + || dynamic_cast(token.get()) + || dynamic_cast(token.get()) + || dynamic_cast(token.get()) + || dynamic_cast(token.get()) + || dynamic_cast(token.get())) { + it--; // unconsume the token + break; // exit the loop + } else { + throw unexpected(**(it-1)); + } + } + if (fully && it != end) { + throw unexpected(**it); + } + if (children.empty()) { + return std::make_shared(Location { template_str, 0 }, std::string()); + } else if (children.size() == 1) { + return std::move(children[0]); + } else { + return std::make_shared(children[0]->location(), std::move(children)); + } + } + +public: + + static std::shared_ptr parse(const std::string& template_str, const Options & options) { + Parser parser(std::make_shared(normalize_newlines(template_str)), options); + auto tokens = parser.tokenize(); + TemplateTokenIterator begin = tokens.begin(); + auto it = begin; + TemplateTokenIterator end = tokens.end(); + return parser.parseTemplate(begin, it, end, /* full= */ true); + } +}; + +static Value simple_function(const std::string & fn_name, const std::vector & params, const std::function &, Value & args)> & fn) { + std::map named_positions; + for (size_t i = 0, n = params.size(); i < n; i++) named_positions[params[i]] = i; + + return Value::callable([=](const std::shared_ptr & context, ArgumentsValue & args) -> Value { + auto args_obj = Value::object(); + std::vector provided_args(params.size()); + for (size_t i = 0, n = args.args.size(); i < n; i++) { + auto & arg = args.args[i]; + if (i < params.size()) { + args_obj.set(params[i], arg); + provided_args[i] = true; + } else { + throw std::runtime_error("Too many positional params for " + fn_name); + } + } + for (auto & [name, value] : args.kwargs) { + auto named_pos_it = named_positions.find(name); + if (named_pos_it == named_positions.end()) { + throw std::runtime_error("Unknown argument " + name + " for function " + fn_name); + } + provided_args[named_pos_it->second] = true; + args_obj.set(name, value); + } + return fn(context, args_obj); + }); +} + +inline std::shared_ptr Context::builtins() { + auto globals = Value::object(); + + globals.set("raise_exception", simple_function("raise_exception", { "message" }, [](const std::shared_ptr &, Value & args) -> Value { + throw std::runtime_error(args.at("message").get()); + })); + globals.set("tojson", simple_function("tojson", { "value", "indent" }, [](const std::shared_ptr &, Value & args) { + return Value(args.at("value").dump(args.get("indent", -1), /* tojson= */ true)); + })); + globals.set("items", simple_function("items", { "object" }, [](const std::shared_ptr &, Value & args) { + auto items = Value::array(); + if (args.contains("object")) { + auto & obj = args.at("object"); + if (obj.is_string()) { + auto json_obj = json::parse(obj.get()); + for (const auto & kv : json_obj.items()) { + items.push_back(Value::array({kv.key(), kv.value()})); + } + } else if (!obj.is_null()) { + for (auto & key : obj.keys()) { + items.push_back(Value::array({key, obj.at(key)})); + } + } + } + return items; + })); + globals.set("last", simple_function("last", { "items" }, [](const std::shared_ptr &, Value & args) { + auto items = args.at("items"); + if (!items.is_array()) throw std::runtime_error("object is not a list"); + if (items.size() == 0) return Value(); + return items.at(items.size() - 1); + })); + globals.set("trim", simple_function("trim", { "text" }, [](const std::shared_ptr &, Value & args) { + auto & text = args.at("text"); + return text.is_null() ? text : Value(strip(text.get())); + })); + globals.set("lower", simple_function("lower", { "text" }, [](const std::shared_ptr &, Value & args) { + auto text = args.at("text"); + if (text.is_null()) return text; + std::string res; + auto str = text.get(); + std::transform(str.begin(), str.end(), std::back_inserter(res), ::tolower); + return Value(res); + })); + globals.set("default", Value::callable([=](const std::shared_ptr &, ArgumentsValue & args) { + args.expectArgs("default", {2, 3}, {0, 1}); + auto & value = args.args[0]; + auto & default_value = args.args[1]; + bool boolean = false; + if (args.args.size() == 3) { + boolean = args.args[2].get(); + } else { + Value bv = args.get_named("boolean"); + if (!bv.is_null()) { + boolean = bv.get(); + } + } + return boolean ? (value.to_bool() ? value : default_value) : value.is_null() ? default_value : value; + })); + auto escape = simple_function("escape", { "text" }, [](const std::shared_ptr &, Value & args) { + return Value(html_escape(args.at("text").get())); + }); + globals.set("e", escape); + globals.set("escape", escape); + globals.set("joiner", simple_function("joiner", { "sep" }, [](const std::shared_ptr &, Value & args) { + auto sep = args.get("sep", ""); + auto first = std::make_shared(true); + return simple_function("", {}, [sep, first](const std::shared_ptr &, const Value &) -> Value { + if (*first) { + *first = false; + return ""; + } + return sep; + }); + return Value(html_escape(args.at("text").get())); + })); + globals.set("count", simple_function("count", { "items" }, [](const std::shared_ptr &, Value & args) { + return Value((int64_t) args.at("items").size()); + })); + globals.set("dictsort", simple_function("dictsort", { "value" }, [](const std::shared_ptr &, Value & args) { + if (args.size() != 1) throw std::runtime_error("dictsort expects exactly 1 argument (TODO: fix implementation)"); + auto & value = args.at("value"); + auto keys = value.keys(); + std::sort(keys.begin(), keys.end()); + auto res = Value::array(); + for (auto & key : keys) { + res.push_back(Value::array({key, value.at(key)})); + } + return res; + })); + globals.set("join", simple_function("join", { "items", "d" }, [](const std::shared_ptr &, Value & args) { + auto do_join = [](Value & items, const std::string & sep) { + std::ostringstream oss; + auto first = true; + for (size_t i = 0, n = items.size(); i < n; ++i) { + if (first) first = false; + else oss << sep; + oss << items.at(i).to_str(); + } + return Value(oss.str()); + }; + auto sep = args.get("d", ""); + if (args.contains("items")) { + auto & items = args.at("items"); + return do_join(items, sep); + } else { + return simple_function("", {"items"}, [sep, do_join](const std::shared_ptr &, Value & args) { + auto & items = args.at("items"); + if (!items.to_bool() || !items.is_array()) throw std::runtime_error("join expects an array for items, got: " + items.dump()); + return do_join(items, sep); + }); + } + })); + globals.set("namespace", Value::callable([=](const std::shared_ptr &, ArgumentsValue & args) { + auto ns = Value::object(); + args.expectArgs("namespace", {0, 0}, {0, (std::numeric_limits::max)()}); + for (auto & [name, value] : args.kwargs) { + ns.set(name, value); + } + return ns; + })); + auto equalto = simple_function("equalto", { "expected", "actual" }, [](const std::shared_ptr &, Value & args) -> Value { + return args.at("actual") == args.at("expected"); + }); + globals.set("equalto", equalto); + globals.set("==", equalto); + globals.set("length", simple_function("length", { "items" }, [](const std::shared_ptr &, Value & args) -> Value { + auto & items = args.at("items"); + return (int64_t) items.size(); + })); + globals.set("safe", simple_function("safe", { "value" }, [](const std::shared_ptr &, Value & args) -> Value { + return args.at("value").to_str(); + })); + globals.set("string", simple_function("string", { "value" }, [](const std::shared_ptr &, Value & args) -> Value { + return args.at("value").to_str(); + })); + globals.set("int", simple_function("int", { "value" }, [](const std::shared_ptr &, Value & args) -> Value { + return args.at("value").to_int(); + })); + globals.set("list", simple_function("list", { "items" }, [](const std::shared_ptr &, Value & args) -> Value { + auto & items = args.at("items"); + if (!items.is_array()) throw std::runtime_error("object is not iterable"); + return items; + })); + globals.set("unique", simple_function("unique", { "items" }, [](const std::shared_ptr &, Value & args) -> Value { + auto & items = args.at("items"); + if (!items.is_array()) throw std::runtime_error("object is not iterable"); + std::unordered_set seen; + auto result = Value::array(); + for (size_t i = 0, n = items.size(); i < n; i++) { + auto pair = seen.insert(items.at(i)); + if (pair.second) { + result.push_back(items.at(i)); + } + } + return result; + })); + auto make_filter = [](const Value & filter, Value & extra_args) -> Value { + return simple_function("", { "value" }, [=](const std::shared_ptr & context, Value & args) { + auto & value = args.at("value"); + ArgumentsValue actual_args; + actual_args.args.emplace_back(value); + for (size_t i = 0, n = extra_args.size(); i < n; i++) { + actual_args.args.emplace_back(extra_args.at(i)); + } + return filter.call(context, actual_args); + }); + }; + // https://jinja.palletsprojects.com/en/3.0.x/templates/#jinja-filters.reject + globals.set("reject", Value::callable([=](const std::shared_ptr & context, ArgumentsValue & args) { + args.expectArgs("reject", {2, (std::numeric_limits::max)()}, {0, 0}); + auto & items = args.args[0]; + auto filter_fn = context->get(args.args[1]); + if (filter_fn.is_null()) throw std::runtime_error("Undefined filter: " + args.args[1].dump()); + + auto filter_args = Value::array(); + for (size_t i = 2, n = args.args.size(); i < n; i++) { + filter_args.push_back(args.args[i]); + } + auto filter = make_filter(filter_fn, filter_args); + + auto res = Value::array(); + for (size_t i = 0, n = items.size(); i < n; i++) { + auto & item = items.at(i); + ArgumentsValue filter_args; + filter_args.args.emplace_back(item); + auto pred_res = filter.call(context, filter_args); + if (!pred_res.to_bool()) { + res.push_back(item); + } + } + return res; + })); + globals.set("map", Value::callable([=](const std::shared_ptr & context, ArgumentsValue & args) { + auto res = Value::array(); + if (args.args.size() == 1 && + ((args.has_named("attribute") && args.kwargs.size() == 1) || (args.has_named("default") && args.kwargs.size() == 2))) { + auto & items = args.args[0]; + auto attr_name = args.get_named("attribute"); + auto default_value = args.get_named("default"); + for (size_t i = 0, n = items.size(); i < n; i++) { + auto & item = items.at(i); + auto attr = item.get(attr_name); + res.push_back(attr.is_null() ? default_value : attr); + } + } else if (args.kwargs.empty() && args.args.size() >= 2) { + auto fn = context->get(args.args[1]); + if (fn.is_null()) throw std::runtime_error("Undefined filter: " + args.args[1].dump()); + ArgumentsValue filter_args { {Value()}, {} }; + for (size_t i = 2, n = args.args.size(); i < n; i++) { + filter_args.args.emplace_back(args.args[i]); + } + for (size_t i = 0, n = args.args[0].size(); i < n; i++) { + auto & item = args.args[0].at(i); + filter_args.args[0] = item; + res.push_back(fn.call(context, filter_args)); + } + } else { + throw std::runtime_error("Invalid or unsupported arguments for map"); + } + return res; + })); + globals.set("indent", simple_function("indent", { "text", "indent", "first" }, [](const std::shared_ptr &, Value & args) { + auto text = args.at("text").get(); + auto first = args.get("first", false); + std::string out; + std::string indent(args.get("indent", 0), ' '); + std::istringstream iss(text); + std::string line; + auto is_first = true; + while (std::getline(iss, line, '\n')) { + auto needs_indent = !is_first || first; + if (is_first) is_first = false; + else out += "\n"; + if (needs_indent) out += indent; + out += line; + } + if (!text.empty() && text.back() == '\n') out += "\n"; + return out; + })); + globals.set("selectattr", Value::callable([=](const std::shared_ptr & context, ArgumentsValue & args) { + args.expectArgs("selectattr", {2, (std::numeric_limits::max)()}, {0, 0}); + auto & items = args.args[0]; + if (items.is_null()) + return Value::array(); + auto attr_name = args.args[1].get(); + + bool has_test = false; + Value test_fn; + ArgumentsValue test_args {{Value()}, {}}; + if (args.args.size() >= 3) { + has_test = true; + test_fn = context->get(args.args[2]); + if (test_fn.is_null()) throw std::runtime_error("Undefined test: " + args.args[2].dump()); + for (size_t i = 3, n = args.args.size(); i < n; i++) { + test_args.args.emplace_back(args.args[i]); + } + test_args.kwargs = args.kwargs; + } + + auto res = Value::array(); + for (size_t i = 0, n = items.size(); i < n; i++) { + auto & item = items.at(i); + auto attr = item.get(attr_name); + if (has_test) { + test_args.args[0] = attr; + if (test_fn.call(context, test_args).to_bool()) { + res.push_back(item); + } + } else { + res.push_back(attr); + } + } + return res; + })); + globals.set("range", Value::callable([=](const std::shared_ptr &, ArgumentsValue & args) { + std::vector startEndStep(3); + std::vector param_set(3); + if (args.args.size() == 1) { + startEndStep[1] = args.args[0].get(); + param_set[1] = true; + } else { + for (size_t i = 0; i < args.args.size(); i++) { + auto & arg = args.args[i]; + auto v = arg.get(); + startEndStep[i] = v; + param_set[i] = true; + } + } + for (auto & [name, value] : args.kwargs) { + size_t i; + if (name == "start") i = 0; + else if (name == "end") i = 1; + else if (name == "step") i = 2; + else throw std::runtime_error("Unknown argument " + name + " for function range"); + + if (param_set[i]) { + throw std::runtime_error("Duplicate argument " + name + " for function range"); + } + startEndStep[i] = value.get(); + param_set[i] = true; + } + if (!param_set[1]) { + throw std::runtime_error("Missing required argument 'end' for function range"); + } + int64_t start = param_set[0] ? startEndStep[0] : 0; + int64_t end = startEndStep[1]; + int64_t step = param_set[2] ? startEndStep[2] : 1; + + auto res = Value::array(); + if (step > 0) { + for (int64_t i = start; i < end; i += step) { + res.push_back(Value(i)); + } + } else { + for (int64_t i = start; i > end; i += step) { + res.push_back(Value(i)); + } + } + return res; + })); + + return std::make_shared(std::move(globals)); +} + +inline std::shared_ptr Context::make(Value && values, const std::shared_ptr & parent) { + return std::make_shared(values.is_null() ? Value::object() : std::move(values), parent); +} + +} // namespace minja diff --git a/common/ngram-cache.cpp b/common/ngram-cache.cpp index a9dfb6714..a057ae45f 100644 --- a/common/ngram-cache.cpp +++ b/common/ngram-cache.cpp @@ -65,13 +65,13 @@ constexpr int draft_min_percent_strict[LLAMA_NGRAM_MAX] = {75, 66, 66, 66}; static llama_token try_draft(common_ngram_cache & nc_static, const common_ngram ngram_static) { common_ngram_cache::iterator part_static_it = nc_static.find(ngram_static); if (part_static_it == nc_static.end()) { - return -1; + return LLAMA_TOKEN_NULL; } const common_ngram_cache_part part_static = part_static_it->second; int max_count_static = 0; int sum_count_static = 0; - llama_token max_token = -1; + llama_token max_token = LLAMA_TOKEN_NULL; for (std::pair token_count_static : part_static) { const llama_token token = token_count_static.first; @@ -85,10 +85,10 @@ static llama_token try_draft(common_ngram_cache & nc_static, const common_ngram } if (sum_count_static < draft_min_sample_size_lax[LLAMA_NGRAM_STATIC-1]) { - return -1; + return LLAMA_TOKEN_NULL; } if (100*max_count_static < draft_min_percent_lax[LLAMA_NGRAM_STATIC-1]*sum_count_static) { - return -1; + return LLAMA_TOKEN_NULL; } return max_token; } @@ -98,9 +98,9 @@ static llama_token try_draft( common_ngram_cache & nc_primary, const std::vector & ngrams_primary, common_ngram_cache_part & part_static, const int * min_sample_size, const int * min_percent) { - llama_token drafted_token = -1; + llama_token drafted_token = LLAMA_TOKEN_NULL; - for (int i = ngrams_primary.size()-1; i >= 0 && drafted_token == -1; --i) { + for (int i = ngrams_primary.size()-1; i >= 0 && drafted_token == LLAMA_TOKEN_NULL; --i) { const common_ngram ngram_primary = ngrams_primary[i]; common_ngram_cache::iterator part_primary_it = nc_primary.find(ngram_primary); @@ -112,7 +112,7 @@ static llama_token try_draft( int max_count_primary = 0; int max_count_static = 0; int sum_count_primary = 0; - llama_token max_token = -1; + llama_token max_token = LLAMA_TOKEN_NULL; for (std::pair token_count_primary : part_primary) { const llama_token token = token_count_primary.first; @@ -154,7 +154,7 @@ void common_ngram_cache_draft( } while ((int) draft.size()-1 < n_draft) { - llama_token drafted_token = -1; + llama_token drafted_token = LLAMA_TOKEN_NULL; const int ngram_start_static = inp_size-LLAMA_NGRAM_STATIC + draft.size()-1; common_ngram ngram_static; @@ -177,17 +177,17 @@ void common_ngram_cache_draft( } ngrams_cd.push_back(ngram_cd); } - if (drafted_token == -1) { + if (drafted_token == LLAMA_TOKEN_NULL) { drafted_token = try_draft(nc_context, ngrams_cd, part_static, draft_min_sample_size_lax, draft_min_percent_lax); } - if (drafted_token == -1) { + if (drafted_token == LLAMA_TOKEN_NULL) { drafted_token = try_draft(nc_dynamic, ngrams_cd, part_static, draft_min_sample_size_strict, draft_min_percent_strict); } - if (drafted_token == -1) { + if (drafted_token == LLAMA_TOKEN_NULL) { drafted_token = try_draft(nc_static, ngram_static); } - if (drafted_token == -1) { + if (drafted_token == LLAMA_TOKEN_NULL) { break; } diff --git a/common/ngram-cache.h b/common/ngram-cache.h index 09c2b0319..dfe012abe 100644 --- a/common/ngram-cache.h +++ b/common/ngram-cache.h @@ -17,13 +17,13 @@ struct common_ngram { common_ngram() { for (int i = 0; i < LLAMA_NGRAM_MAX; ++i) { - tokens[i] = -1; + tokens[i] = LLAMA_TOKEN_NULL; } } common_ngram(const llama_token * input, const int ngram_size) { for (int i = 0; i < LLAMA_NGRAM_MAX; ++i) { - tokens[i] = i < ngram_size ? input[i] : -1; + tokens[i] = i < ngram_size ? input[i] : LLAMA_TOKEN_NULL; } } diff --git a/common/sampling.cpp b/common/sampling.cpp index e83a971c7..7241ac321 100644 --- a/common/sampling.cpp +++ b/common/sampling.cpp @@ -113,7 +113,10 @@ struct common_sampler { void set_logits(struct llama_context * ctx, int idx) { const auto * logits = llama_get_logits_ith(ctx, idx); - const int n_vocab = llama_n_vocab(llama_get_model(ctx)); + const llama_model * model = llama_get_model(ctx); + const llama_vocab * vocab = llama_model_get_vocab(model); + + const int n_vocab = llama_vocab_n_tokens(vocab); cur.resize(n_vocab); @@ -142,13 +145,15 @@ std::string common_params_sampling::print() const { } struct common_sampler * common_sampler_init(const struct llama_model * model, const struct common_params_sampling & params) { + const llama_vocab * vocab = llama_model_get_vocab(model); + llama_sampler_chain_params lparams = llama_sampler_chain_default_params(); lparams.no_perf = params.no_perf; auto * result = new common_sampler { /* .params = */ params, - /* .grmr = */ llama_sampler_init_grammar(model, params.grammar.c_str(), "root"), + /* .grmr = */ llama_sampler_init_grammar(vocab, params.grammar.c_str(), "root"), /* .chain = */ llama_sampler_chain_init(lparams), /* .prev = */ ring_buffer(std::max(32, params.n_prev)), /* .cur = */ {}, @@ -157,7 +162,7 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, co llama_sampler_chain_add(result->chain, llama_sampler_init_logit_bias( - llama_n_vocab(model), + llama_vocab_n_tokens(vocab), params.logit_bias.size(), params.logit_bias.data())); @@ -172,7 +177,7 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, co c_breakers.push_back(str.c_str()); } - llama_sampler_chain_add(result->chain, llama_sampler_init_dry (model, params.dry_multiplier, params.dry_base, params.dry_allowed_length, params.dry_penalty_last_n, c_breakers.data(), c_breakers.size())); + llama_sampler_chain_add(result->chain, llama_sampler_init_dry (vocab, llama_model_n_ctx_train(model), params.dry_multiplier, params.dry_base, params.dry_allowed_length, params.dry_penalty_last_n, c_breakers.data(), c_breakers.size())); } break; case COMMON_SAMPLER_TYPE_TOP_K: @@ -194,7 +199,7 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, co llama_sampler_chain_add(result->chain, llama_sampler_init_temp_ext (params.temp, params.dynatemp_range, params.dynatemp_exponent)); break; case COMMON_SAMPLER_TYPE_INFILL: - llama_sampler_chain_add(result->chain, llama_sampler_init_infill (model)); + llama_sampler_chain_add(result->chain, llama_sampler_init_infill (vocab)); break; case COMMON_SAMPLER_TYPE_PENALTIES: llama_sampler_chain_add(result->chain, llama_sampler_init_penalties(params.penalty_last_n, params.penalty_repeat, params.penalty_freq, params.penalty_present)); @@ -206,7 +211,7 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, co llama_sampler_chain_add(result->chain, llama_sampler_init_dist(params.seed)); } else if (params.mirostat == 1) { llama_sampler_chain_add(result->chain, llama_sampler_init_temp(params.temp)); - llama_sampler_chain_add(result->chain, llama_sampler_init_mirostat(llama_n_vocab(model), params.seed, params.mirostat_tau, params.mirostat_eta, 100)); + llama_sampler_chain_add(result->chain, llama_sampler_init_mirostat(llama_vocab_n_tokens(vocab), params.seed, params.mirostat_tau, params.mirostat_eta, 100)); } else if (params.mirostat == 2) { llama_sampler_chain_add(result->chain, llama_sampler_init_temp(params.temp)); llama_sampler_chain_add(result->chain, llama_sampler_init_mirostat_v2(params.seed, params.mirostat_tau, params.mirostat_eta)); diff --git a/common/speculative.cpp b/common/speculative.cpp index 3fcbb0020..318e96ea3 100644 --- a/common/speculative.cpp +++ b/common/speculative.cpp @@ -79,10 +79,13 @@ bool common_speculative_are_compatible( const struct llama_model * model_tgt = llama_get_model(ctx_tgt); const struct llama_model * model_dft = llama_get_model(ctx_dft); - const bool vocab_type_tgt = llama_vocab_type(model_tgt); + const struct llama_vocab * vocab_tgt = llama_model_get_vocab(model_tgt); + const struct llama_vocab * vocab_dft = llama_model_get_vocab(model_dft); + + const bool vocab_type_tgt = llama_vocab_type(vocab_tgt); LOG_DBG("%s: vocab_type tgt: %d\n", __func__, vocab_type_tgt); - const bool vocab_type_dft = llama_vocab_type(model_dft); + const bool vocab_type_dft = llama_vocab_type(vocab_dft); LOG_DBG("%s: vocab_type dft: %d\n", __func__, vocab_type_dft); if (vocab_type_tgt != vocab_type_dft) { @@ -91,34 +94,34 @@ bool common_speculative_are_compatible( return false; } - if (llama_add_bos_token(model_tgt) != llama_add_bos_token(model_dft) || - llama_add_eos_token(model_tgt) != llama_add_eos_token(model_dft) || - llama_token_bos(model_tgt) != llama_token_bos(model_dft) || - llama_token_eos(model_tgt) != llama_token_eos(model_dft)) { - LOG_ERR("%s: draft model special tokens must match target model to use speculation\n", __func__); - LOG_ERR("%s: tgt: bos = %d (%d), eos = %d (%d)\n", __func__, llama_token_bos(model_tgt), llama_add_bos_token(model_tgt), llama_token_eos(model_tgt), llama_add_eos_token(model_tgt)); - LOG_ERR("%s: dft: bos = %d (%d), eos = %d (%d)\n", __func__, llama_token_bos(model_dft), llama_add_bos_token(model_dft), llama_token_eos(model_dft), llama_add_eos_token(model_dft)); + if (llama_vocab_get_add_bos(vocab_tgt) != llama_vocab_get_add_bos(vocab_dft) || + llama_vocab_get_add_eos(vocab_tgt) != llama_vocab_get_add_eos(vocab_dft) || + llama_vocab_bos(vocab_tgt) != llama_vocab_bos(vocab_dft) || + llama_vocab_eos(vocab_tgt) != llama_vocab_eos(vocab_dft)) { + LOG_ERR("%s: draft vocab special tokens must match target vocab to use speculation\n", __func__); + LOG_ERR("%s: tgt: bos = %d (%d), eos = %d (%d)\n", __func__, llama_vocab_bos(vocab_tgt), llama_vocab_get_add_bos(vocab_tgt), llama_vocab_eos(vocab_tgt), llama_vocab_get_add_eos(vocab_tgt)); + LOG_ERR("%s: dft: bos = %d (%d), eos = %d (%d)\n", __func__, llama_vocab_bos(vocab_dft), llama_vocab_get_add_bos(vocab_dft), llama_vocab_eos(vocab_dft), llama_vocab_get_add_eos(vocab_dft)); return false; } { - const int n_vocab_tgt = llama_n_vocab(model_tgt); - const int n_vocab_dft = llama_n_vocab(model_dft); + const int n_vocab_tgt = llama_vocab_n_tokens(vocab_tgt); + const int n_vocab_dft = llama_vocab_n_tokens(vocab_dft); const int vocab_diff = std::abs(n_vocab_tgt - n_vocab_dft); if (vocab_diff > SPEC_VOCAB_MAX_SIZE_DIFFERENCE) { LOG_ERR("%s: draft model vocab must closely match target model to use speculation but " "target vocab size %d does not match draft vocab size %d - difference %d, max allowed %d\n", - __func__, n_vocab_tgt, llama_n_vocab(model_dft), vocab_diff, SPEC_VOCAB_MAX_SIZE_DIFFERENCE); + __func__, n_vocab_tgt, llama_vocab_n_tokens(vocab_dft), vocab_diff, SPEC_VOCAB_MAX_SIZE_DIFFERENCE); return false; } for (int i = SPEC_VOCAB_CHECK_START_TOKEN_ID; i < std::min(n_vocab_tgt, n_vocab_dft); ++i) { - const char * token_text_tgt = llama_token_get_text(model_tgt, i); - const char * token_text_dft = llama_token_get_text(model_dft, i); + const char * token_text_tgt = llama_vocab_get_text(vocab_tgt, i); + const char * token_text_dft = llama_vocab_get_text(vocab_dft, i); if (std::strcmp(token_text_tgt, token_text_dft) != 0) { - LOG_ERR("%s: draft model vocab must match target model to use speculation but " + LOG_ERR("%s: draft vocab vocab must match target vocab to use speculation but " "token %d content differs - target '%s', draft '%s'\n", __func__, i, common_token_to_piece(ctx_tgt, i).c_str(), common_token_to_piece(ctx_dft, i).c_str()); diff --git a/convert_hf_to_gguf.py b/convert_hf_to_gguf.py index 01b58f976..63b54a9cf 100755 --- a/convert_hf_to_gguf.py +++ b/convert_hf_to_gguf.py @@ -326,6 +326,7 @@ class Model: gguf.MODEL_TENSOR.TIME_MIX_W2, gguf.MODEL_TENSOR.TIME_MIX_DECAY_W1, gguf.MODEL_TENSOR.TIME_MIX_DECAY_W2, + gguf.MODEL_TENSOR.TIME_MIX_LERP_FUSED, gguf.MODEL_TENSOR.POSNET_NORM1, gguf.MODEL_TENSOR.POSNET_NORM2, ) @@ -477,6 +478,11 @@ class Model: return modelcls return func + @classmethod + def print_registered_models(cls): + for name in sorted(cls._model_classes.keys()): + logger.error(f"- {name}") + @classmethod def from_model_architecture(cls, arch: str) -> type[Model]: try: @@ -690,6 +696,9 @@ class Model: if chkhsh == "877081d19cf6996e2c4ff0e1236341e9b7bde288f5311a56a937f0afbbb3aeb5": # ref: https://huggingface.co/deepseek-ai/DeepSeek-V3 res = "deepseek-v3" + if chkhsh == "b3f499bb4255f8ca19fccd664443283318f2fd2414d5e0b040fbdd0cc195d6c5": + # ref: https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B + res = "deepseek-r1-qwen" if res is None: logger.warning("\n") @@ -2562,6 +2571,63 @@ class Phi3MiniModel(Model): yield (self.format_tensor_name(gguf.MODEL_TENSOR.ROPE_FACTORS_SHORT), torch.tensor(short_factors, dtype=torch.float32)) +@Model.register("PhiMoEForCausalLM") +class PhiMoeModel(Phi3MiniModel): + model_arch = gguf.MODEL_ARCH.PHIMOE + + _experts: list[dict[str, Tensor]] | None = None + + def set_gguf_parameters(self): + super().set_gguf_parameters() + self.gguf_writer.add_expert_used_count(self.hparams["num_experts_per_tok"]) + self.gguf_writer.add_expert_count(self.hparams["num_local_experts"]) + + def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]: + # process the experts separately + if name.find("block_sparse_moe.experts") != -1: + n_experts = self.hparams["num_local_experts"] + assert bid is not None + + if self._experts is None: + self._experts = [{} for _ in range(self.block_count)] + + self._experts[bid][name] = data_torch + + if len(self._experts[bid]) >= n_experts * 3: + tensors: list[tuple[str, Tensor]] = [] + + # merge the experts into a single 3d tensor + for w_name in ["w1", "w2", "w3"]: + datas: list[Tensor] = [] + + for xid in range(n_experts): + ename = f"model.layers.{bid}.block_sparse_moe.experts.{xid}.{w_name}.weight" + datas.append(self._experts[bid][ename]) + del self._experts[bid][ename] + + data_torch = torch.stack(datas, dim=0) + + merged_name = f"model.layers.{bid}.block_sparse_moe.experts.{w_name}.weight" + + new_name = self.map_tensor_name(merged_name) + + tensors.append((new_name, data_torch)) + return tensors + else: + return [] + + return [(self.map_tensor_name(name), data_torch)] + + def prepare_tensors(self): + super().prepare_tensors() + + if self._experts is not None: + # flatten `list[dict[str, Tensor]]` into `list[str]` + experts = [k for d in self._experts for k in d.keys()] + if len(experts) > 0: + raise ValueError(f"Unprocessed experts: {experts}") + + @Model.register("PlamoForCausalLM") class PlamoModel(Model): model_arch = gguf.MODEL_ARCH.PLAMO @@ -2819,6 +2885,66 @@ class InternLM2Model(Model): return [(self.map_tensor_name(name), data_torch)] +@Model.register("InternLM3ForCausalLM") +class InternLM3Model(Model): + model_arch = gguf.MODEL_ARCH.LLAMA + + def set_vocab(self): + tokens, scores, toktypes = self._create_vocab_sentencepiece() + + self.gguf_writer.add_tokenizer_model("llama") + self.gguf_writer.add_tokenizer_pre("default") + self.gguf_writer.add_token_list(tokens) + self.gguf_writer.add_token_scores(scores) + self.gguf_writer.add_token_types(toktypes) + + special_vocab = gguf.SpecialVocab(self.dir_model, n_vocab=len(tokens)) + + tokenizer_config_file = self.dir_model / 'tokenizer_config.json' + if tokenizer_config_file.is_file(): + with open(tokenizer_config_file, "r", encoding="utf-8") as f: + tokenizer_config_json = json.load(f) + if "add_prefix_space" in tokenizer_config_json: + self.gguf_writer.add_add_space_prefix(tokenizer_config_json["add_prefix_space"]) + + if "added_tokens_decoder" in tokenizer_config_json: + for token_id, token_data in tokenizer_config_json["added_tokens_decoder"].items(): + if token_data.get("special"): + token_id = int(token_id) + token = token_data["content"] + special_vocab._set_special_token(token, token_id) + # update eos token + if token == '<|im_end|>' and "eos" in special_vocab.special_token_ids: + special_vocab.special_token_ids["eos"] = token_id + + special_vocab.add_to_gguf(self.gguf_writer) + + def set_gguf_parameters(self): + super().set_gguf_parameters() + hparams = self.hparams + self.gguf_writer.add_vocab_size(hparams["vocab_size"]) + + if "head_dim" in hparams: + rope_dim = hparams["head_dim"] + else: + rope_dim = hparams["hidden_size"] // hparams["num_attention_heads"] + self.gguf_writer.add_rope_dimension_count(rope_dim) + + if self.hparams.get("rope_scaling") is not None and "factor" in self.hparams["rope_scaling"]: + if self.hparams["rope_scaling"].get("type") == "linear" or self.hparams["rope_scaling"].get("rope_type") == "linear": + self.gguf_writer.add_rope_scaling_type(gguf.RopeScalingType.LINEAR) + self.gguf_writer.add_rope_scaling_factor(self.hparams["rope_scaling"]["factor"]) + + def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]: + n_head = self.hparams["num_attention_heads"] + n_kv_head = self.hparams.get("num_key_value_heads") + if name.endswith(("q_proj.weight", "q_proj.bias")): + data_torch = LlamaModel.permute(data_torch, n_head, n_head) + if name.endswith(("k_proj.weight", "k_proj.bias")): + data_torch = LlamaModel.permute(data_torch, n_head, n_kv_head) + return [(self.map_tensor_name(name), data_torch)] + + @Model.register("BertModel", "BertForMaskedLM", "CamembertModel") class BertModel(Model): model_arch = gguf.MODEL_ARCH.BERT @@ -3259,6 +3385,8 @@ class Rwkv6Model(Model): # required by llama.cpp, unused self.gguf_writer.add_head_count(0) + lerp_weights: dict[int, dict[str, Tensor]] = {} + def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]: new_name = self.map_tensor_name(name) @@ -3274,14 +3402,84 @@ class Rwkv6Model(Model): if new_name.endswith("time_mix_decay.weight") or "lerp" in new_name: data_torch = data_torch.squeeze() - rescale_every_n_layers = self.hparams["rescale_every"] - if rescale_every_n_layers > 0: - if new_name.endswith("time_mix_output.weight") or new_name.endswith("channel_mix_value.weight"): - data_torch = data_torch.div_(2 ** int(bid // rescale_every_n_layers)) + try: + rescale_every_n_layers = self.hparams["rescale_every"] + if rescale_every_n_layers > 0: + if new_name.endswith("time_mix_output.weight") or new_name.endswith("channel_mix_value.weight"): + data_torch = data_torch.div_(2 ** int(bid // rescale_every_n_layers)) + except KeyError: + pass + + # concat time_mix_lerp weights to reduce some cpu overhead + # also reduces the number of tensors in the model + if bid is not None and "time_mix_lerp" in new_name and "time_mix_lerp_x" not in new_name: + try: + self.lerp_weights[bid][new_name] = data_torch + except KeyError: + self.lerp_weights[bid] = {new_name: data_torch} + if all(f"blk.{bid}.time_mix_lerp_{i}.weight" in self.lerp_weights[bid].keys() for i in ["w", "k", "v", "r", "g"]): + new_name = f"blk.{bid}.time_mix_lerp_fused.weight" + data = torch.stack([self.lerp_weights[bid][f"blk.{bid}.time_mix_lerp_{i}.weight"].unsqueeze(0) for i in ["w", "k", "v", "r", "g"]], dim=0).unsqueeze(1) + yield (new_name, data) + return yield (new_name, data_torch) +@Model.register("RWKV6Qwen2ForCausalLM") +class RWKV6Qwen2Model(Rwkv6Model): + model_arch = gguf.MODEL_ARCH.RWKV6QWEN2 + + def set_vocab(self): + try: + self._set_vocab_sentencepiece() + except FileNotFoundError: + self._set_vocab_gpt2() + + def set_gguf_parameters(self): + block_count = self.hparams["num_hidden_layers"] + num_attention_heads = self.hparams["num_attention_heads"] + num_key_value_heads = self.hparams["num_key_value_heads"] + hidden_size = self.hparams["hidden_size"] + head_size = hidden_size // num_attention_heads + rms_norm_eps = self.hparams["rms_norm_eps"] + intermediate_size = self.hparams["intermediate_size"] + time_mix_extra_dim = 64 if hidden_size >= 4096 else 32 + time_decay_extra_dim = 128 if hidden_size >= 4096 else 64 + + # RWKV isn't context limited + self.gguf_writer.add_context_length(1048576) + self.gguf_writer.add_embedding_length(hidden_size) + self.gguf_writer.add_block_count(block_count) + self.gguf_writer.add_wkv_head_size(head_size) + self.gguf_writer.add_time_mix_extra_dim(time_mix_extra_dim) + self.gguf_writer.add_time_decay_extra_dim(time_decay_extra_dim) + self.gguf_writer.add_feed_forward_length(intermediate_size) + self.gguf_writer.add_file_type(self.ftype) + + # special parameters for time_mixing in RWKV6QWEN2 + self.gguf_writer.add_layer_norm_rms_eps(rms_norm_eps) + self.gguf_writer.add_token_shift_count(1) + # RWKV6QWEN2 use grouped key/value like GQA + self.gguf_writer.add_head_count_kv(num_key_value_heads) + + # required by llama.cpp, unused + self.gguf_writer.add_head_count(0) + + def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]: + for new_name, data in super().modify_tensors(data_torch, name, bid): + if "time_mix_w1" in new_name or "time_mix_w2" in new_name: + data = data.view(5, -1, data.shape[-1]) + # rwkv6qwen2 has a different order of rkvwg instead of the original wkvrg + # permute them here to avoid code changes + data = torch.stack([data[3], data[1], data[2], data[0], data[4]], dim=0).view(-1, data.shape[-1]) + if "w2" in new_name: + data = data.view(5, -1, data.shape[-1]) + yield (new_name, data) + continue + yield (new_name, data) + + @Model.register("MambaForCausalLM", "MambaLMHeadModel", "FalconMambaForCausalLM") class MambaModel(Model): model_arch = gguf.MODEL_ARCH.MAMBA @@ -4799,6 +4997,7 @@ def parse_args() -> argparse.Namespace: parser.add_argument( "model", type=Path, help="directory containing model file", + nargs="?", ) parser.add_argument( "--use-temp-file", action="store_true", @@ -4836,8 +5035,15 @@ def parse_args() -> argparse.Namespace: "--metadata", type=Path, help="Specify the path for an authorship metadata override file" ) + parser.add_argument( + "--print-supported-models", action="store_true", + help="Print the supported models" + ) - return parser.parse_args() + args = parser.parse_args() + if not args.print_supported_models and args.model is None: + parser.error("the following arguments are required: model") + return args def split_str_to_n_bytes(split_str: str) -> int: @@ -4861,6 +5067,11 @@ def split_str_to_n_bytes(split_str: str) -> int: def main() -> None: args = parse_args() + if args.print_supported_models: + logger.error("Supported models:") + Model.print_registered_models() + sys.exit(0) + if args.verbose: logging.basicConfig(level=logging.DEBUG) else: diff --git a/convert_hf_to_gguf_update.py b/convert_hf_to_gguf_update.py index 56edc64a7..cea34413f 100755 --- a/convert_hf_to_gguf_update.py +++ b/convert_hf_to_gguf_update.py @@ -65,49 +65,50 @@ else: # TODO: add models here, base models preferred models = [ - {"name": "llama-spm", "tokt": TOKENIZER_TYPE.SPM, "repo": "https://huggingface.co/meta-llama/Llama-2-7b-hf", }, - {"name": "llama-bpe", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/meta-llama/Meta-Llama-3-8B", }, - {"name": "phi-3", "tokt": TOKENIZER_TYPE.SPM, "repo": "https://huggingface.co/microsoft/Phi-3-mini-4k-instruct", }, - {"name": "deepseek-llm", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/deepseek-ai/deepseek-llm-7b-base", }, - {"name": "deepseek-coder", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/deepseek-ai/deepseek-coder-6.7b-base", }, - {"name": "falcon", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/tiiuae/falcon-7b", }, - {"name": "bert-bge", "tokt": TOKENIZER_TYPE.WPM, "repo": "https://huggingface.co/BAAI/bge-small-en-v1.5", }, - {"name": "falcon3", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/tiiuae/Falcon3-7B-Base", }, - {"name": "bert-bge-large", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/BAAI/bge-large-zh-v1.5", }, - {"name": "mpt", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/mosaicml/mpt-7b", }, - {"name": "starcoder", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/bigcode/starcoder2-3b", }, - {"name": "gpt-2", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/openai-community/gpt2", }, - {"name": "stablelm2", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/stabilityai/stablelm-2-zephyr-1_6b", }, - {"name": "refact", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/smallcloudai/Refact-1_6-base", }, - {"name": "command-r", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/CohereForAI/c4ai-command-r-v01", }, - {"name": "qwen2", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/Qwen/Qwen1.5-7B", }, - {"name": "olmo", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/allenai/OLMo-1.7-7B-hf", }, - {"name": "dbrx", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/databricks/dbrx-base", }, - {"name": "jina-v1-en", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-reranker-v1-tiny-en", }, - {"name": "jina-v2-en", "tokt": TOKENIZER_TYPE.WPM, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-en", }, # WPM! - {"name": "jina-v2-es", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-es", }, - {"name": "jina-v2-de", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-de", }, - {"name": "smaug-bpe", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/abacusai/Smaug-Llama-3-70B-Instruct", }, - {"name": "poro-chat", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LumiOpen/Poro-34B-chat", }, - {"name": "jina-v2-code", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-code", }, - {"name": "viking", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LumiOpen/Viking-7B", }, # Also used for Viking 13B and 33B - {"name": "gemma", "tokt": TOKENIZER_TYPE.SPM, "repo": "https://huggingface.co/google/gemma-2b", }, - {"name": "gemma-2", "tokt": TOKENIZER_TYPE.SPM, "repo": "https://huggingface.co/google/gemma-2-9b", }, - {"name": "jais", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/core42/jais-13b", }, - {"name": "t5", "tokt": TOKENIZER_TYPE.UGM, "repo": "https://huggingface.co/google-t5/t5-small", }, - {"name": "codeshell", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/WisdomShell/CodeShell-7B", }, - {"name": "tekken", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/mistralai/Mistral-Nemo-Base-2407", }, - {"name": "smollm", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/HuggingFaceTB/SmolLM-135M", }, - {'name': "bloom", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/bigscience/bloom", }, - {'name': "gpt3-finnish", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/TurkuNLP/gpt3-finnish-small", }, - {"name": "exaone", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LGAI-EXAONE/EXAONE-3.0-7.8B-Instruct", }, - {"name": "phi-2", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/microsoft/phi-2", }, - {"name": "chameleon", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/facebook/chameleon-7b", }, - {"name": "minerva-7b", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/sapienzanlp/Minerva-7B-base-v1.0", }, - {"name": "roberta-bpe", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/sentence-transformers/stsb-roberta-base"}, - {"name": "gigachat", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/ai-sage/GigaChat-20B-A3B-instruct"}, - {"name": "megrez", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/Infinigence/Megrez-3B-Instruct"}, - {"name": "deepseek-v3", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/deepseek-ai/DeepSeek-V3"}, + {"name": "llama-spm", "tokt": TOKENIZER_TYPE.SPM, "repo": "https://huggingface.co/meta-llama/Llama-2-7b-hf", }, + {"name": "llama-bpe", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/meta-llama/Meta-Llama-3-8B", }, + {"name": "phi-3", "tokt": TOKENIZER_TYPE.SPM, "repo": "https://huggingface.co/microsoft/Phi-3-mini-4k-instruct", }, + {"name": "deepseek-llm", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/deepseek-ai/deepseek-llm-7b-base", }, + {"name": "deepseek-coder", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/deepseek-ai/deepseek-coder-6.7b-base", }, + {"name": "falcon", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/tiiuae/falcon-7b", }, + {"name": "bert-bge", "tokt": TOKENIZER_TYPE.WPM, "repo": "https://huggingface.co/BAAI/bge-small-en-v1.5", }, + {"name": "falcon3", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/tiiuae/Falcon3-7B-Base", }, + {"name": "bert-bge-large", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/BAAI/bge-large-zh-v1.5", }, + {"name": "mpt", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/mosaicml/mpt-7b", }, + {"name": "starcoder", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/bigcode/starcoder2-3b", }, + {"name": "gpt-2", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/openai-community/gpt2", }, + {"name": "stablelm2", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/stabilityai/stablelm-2-zephyr-1_6b", }, + {"name": "refact", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/smallcloudai/Refact-1_6-base", }, + {"name": "command-r", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/CohereForAI/c4ai-command-r-v01", }, + {"name": "qwen2", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/Qwen/Qwen1.5-7B", }, + {"name": "olmo", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/allenai/OLMo-1.7-7B-hf", }, + {"name": "dbrx", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/databricks/dbrx-base", }, + {"name": "jina-v1-en", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-reranker-v1-tiny-en", }, + {"name": "jina-v2-en", "tokt": TOKENIZER_TYPE.WPM, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-en", }, # WPM! + {"name": "jina-v2-es", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-es", }, + {"name": "jina-v2-de", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-de", }, + {"name": "smaug-bpe", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/abacusai/Smaug-Llama-3-70B-Instruct", }, + {"name": "poro-chat", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LumiOpen/Poro-34B-chat", }, + {"name": "jina-v2-code", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-code", }, + {"name": "viking", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LumiOpen/Viking-7B", }, # Also used for Viking 13B and 33B + {"name": "gemma", "tokt": TOKENIZER_TYPE.SPM, "repo": "https://huggingface.co/google/gemma-2b", }, + {"name": "gemma-2", "tokt": TOKENIZER_TYPE.SPM, "repo": "https://huggingface.co/google/gemma-2-9b", }, + {"name": "jais", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/core42/jais-13b", }, + {"name": "t5", "tokt": TOKENIZER_TYPE.UGM, "repo": "https://huggingface.co/google-t5/t5-small", }, + {"name": "codeshell", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/WisdomShell/CodeShell-7B", }, + {"name": "tekken", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/mistralai/Mistral-Nemo-Base-2407", }, + {"name": "smollm", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/HuggingFaceTB/SmolLM-135M", }, + {'name': "bloom", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/bigscience/bloom", }, + {'name': "gpt3-finnish", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/TurkuNLP/gpt3-finnish-small", }, + {"name": "exaone", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LGAI-EXAONE/EXAONE-3.0-7.8B-Instruct", }, + {"name": "phi-2", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/microsoft/phi-2", }, + {"name": "chameleon", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/facebook/chameleon-7b", }, + {"name": "minerva-7b", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/sapienzanlp/Minerva-7B-base-v1.0", }, + {"name": "roberta-bpe", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/sentence-transformers/stsb-roberta-base"}, + {"name": "gigachat", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/ai-sage/GigaChat-20B-A3B-instruct"}, + {"name": "megrez", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/Infinigence/Megrez-3B-Instruct"}, + {"name": "deepseek-v3", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/deepseek-ai/DeepSeek-V3"}, + {"name": "deepseek-r1-qwen", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B"}, ] diff --git a/convert_lora_to_gguf.py b/convert_lora_to_gguf.py index ed1014cae..6dea14a23 100755 --- a/convert_lora_to_gguf.py +++ b/convert_lora_to_gguf.py @@ -226,6 +226,9 @@ def get_base_tensor_name(lora_tensor_name: str) -> str: base_name = lora_tensor_name.replace("base_model.model.", "") base_name = base_name.replace(".lora_A.weight", ".weight") base_name = base_name.replace(".lora_B.weight", ".weight") + # models produced by mergekit-extract-lora have token embeddings in the adapter + base_name = base_name.replace(".lora_embedding_A", ".weight") + base_name = base_name.replace(".lora_embedding_B", ".weight") return base_name @@ -260,6 +263,10 @@ def parse_args() -> argparse.Namespace: "--base", type=Path, help="directory containing Hugging Face model config files (config.json, tokenizer.json) for the base model that the adapter is based on - only config is needed, actual model weights are not required. If base model is unspecified, it will be loaded from Hugging Face hub based on the adapter config", ) + parser.add_argument( + "--base-model-id", type=str, + help="the model ID of the base model, if it is not available locally or in the adapter config. If specified, it will ignore --base and load the base model config from the Hugging Face hub (Example: 'meta-llama/Llama-3.2-1B-Instruct')", + ) parser.add_argument( "lora_path", type=Path, help="directory containing Hugging Face PEFT LoRA config (adapter_model.json) and weights (adapter_model.safetensors or adapter_model.bin)", @@ -290,6 +297,7 @@ if __name__ == '__main__': dir_base_model: Path | None = args.base dir_lora: Path = args.lora_path + base_model_id: str | None = args.base_model_id lora_config = dir_lora / "adapter_config.json" input_model = dir_lora / "adapter_model.safetensors" @@ -313,7 +321,10 @@ if __name__ == '__main__': lparams: dict[str, Any] = json.load(f) # load base model - if dir_base_model is None: + if base_model_id is not None: + logger.info(f"Loading base model from Hugging Face: {base_model_id}") + hparams = load_hparams_from_hf(base_model_id) + elif dir_base_model is None: if "base_model_name_or_path" in lparams: model_id = lparams["base_model_name_or_path"] logger.info(f"Loading base model from Hugging Face: {model_id}") @@ -371,11 +382,16 @@ if __name__ == '__main__': if self.lazy: tensor = LazyTorchTensor.from_eager(tensor) base_name = get_base_tensor_name(name) - is_lora_a = ".lora_A.weight" in name - is_lora_b = ".lora_B.weight" in name + # note: mergekit-extract-lora also adds token embeddings to the adapter + is_lora_a = ".lora_A.weight" in name or ".lora_embedding_A" in name + is_lora_b = ".lora_B.weight" in name or ".lora_embedding_B" in name if not is_lora_a and not is_lora_b: if ".base_layer.weight" in name: continue + # mergekit-extract-lora add these layernorm to the adapter, we need to keep them + if "_layernorm" in name or ".norm" in name: + yield (base_name, tensor) + continue logger.error(f"Unexpected name '{name}': Not a lora_A or lora_B tensor") if ".embed_tokens.weight" in name or ".lm_head.weight" in name: logger.error("Embeddings is present in the adapter. This can be due to new tokens added during fine tuning") @@ -407,9 +423,21 @@ if __name__ == '__main__': if name == "lm_head.weight" and len(dest) == 0: raise ValueError("lm_head is present in adapter, but is ignored in base model") for dest_name, dest_data in dest: + # mergekit-extract-lora add these layernorm to the adapter + if "_norm" in dest_name: + assert dest_data.dim() == 1 + yield (dest_name, dest_data) + continue + + # otherwise, we must get the lora_A and lora_B tensors assert isinstance(dest_data, LoraTorchTensor) lora_a, lora_b = dest_data.get_lora_A_B() + # note: mergekit-extract-lora flip and transpose A and B + # here we only need to transpose token_embd.lora_a, see llm_build_inp_embd() + if "token_embd.weight" in dest_name: + lora_a = lora_a.T + yield (dest_name + ".lora_a", lora_a) yield (dest_name + ".lora_b", lora_b) diff --git a/docs/backend/SYCL.md b/docs/backend/SYCL.md index 8d8312e91..89ddbd669 100644 --- a/docs/backend/SYCL.md +++ b/docs/backend/SYCL.md @@ -133,7 +133,7 @@ The docker build option is currently limited to *intel GPU* targets. ### Build image ```sh # Using FP16 -docker build -t llama-cpp-sycl --build-arg="GGML_SYCL_F16=ON" -f .devops/llama-cli-intel.Dockerfile . +docker build -t llama-cpp-sycl --build-arg="GGML_SYCL_F16=ON" --target light -f .devops/intel.Dockerfile . ``` *Notes*: diff --git a/docs/build.md b/docs/build.md index 84019b204..dd6495028 100644 --- a/docs/build.md +++ b/docs/build.md @@ -127,6 +127,8 @@ For detailed info, please refer to [llama.cpp for SYCL](./backend/SYCL.md). This provides GPU acceleration using an NVIDIA GPU. Make sure to have the CUDA toolkit installed. You can download it from your Linux distro's package manager (e.g. `apt install nvidia-cuda-toolkit`) or from the [NVIDIA developer site](https://developer.nvidia.com/cuda-downloads). +If you are using Fedora (using Fedora Workstation, or an 'Atomic' variant such as Silverblue), or would like to set up CUDA in a toolbox, please consider our [Fedora CUDA guide](./cuda-fedora.md). Unfortunately, the process is not as simple as one might expect. + - Using `CMake`: ```bash @@ -284,7 +286,7 @@ You don't need to install Vulkan SDK. It will be installed inside the container. ```sh # Build the image -docker build -t llama-cpp-vulkan -f .devops/llama-cli-vulkan.Dockerfile . +docker build -t llama-cpp-vulkan --target light -f .devops/vulkan.Dockerfile . # Then, use it: docker run -it --rm -v "$(pwd):/app:Z" --device /dev/dri/renderD128:/dev/dri/renderD128 --device /dev/dri/card1:/dev/dri/card1 llama-cpp-vulkan -m "/app/models/YOUR_MODEL_FILE" -p "Building a website can be done in 10 simple steps:" -n 400 -e -ngl 33 diff --git a/docs/cuda-fedora.md b/docs/cuda-fedora.md new file mode 100644 index 000000000..b993386c8 --- /dev/null +++ b/docs/cuda-fedora.md @@ -0,0 +1,317 @@ +# Setting Up CUDA on Fedora + +In this guide we setup [Nvidia CUDA](https://docs.nvidia.com/cuda/) in a toolbox container. This guide is applicable for: +- [Fedora Workstation](https://fedoraproject.org/workstation/) +- [Atomic Desktops for Fedora](https://fedoraproject.org/atomic-desktops/) +- [Fedora Spins](https://fedoraproject.org/spins) +- [Other Distributions](https://containertoolbx.org/distros/), including `Red Hat Enterprise Linux >= 8.`, `Arch Linux`, and `Ubuntu`. + + +## Table of Contents + +- [Prerequisites](#prerequisites) +- [Monitoring NVIDIA CUDA Repositories](#monitoring-nvidia-cuda-repositories) +- [Using the Fedora 39 CUDA Repository](#using-the-fedora-39-cuda-repository) +- [Creating a Fedora Toolbox Environment](#creating-a-fedora-toolbox-environment) +- [Installing Essential Development Tools](#installing-essential-development-tools) +- [Adding the CUDA Repository](#adding-the-cuda-repository) +- [Installing `nvidia-driver-libs`](#installing-nvidia-driver-libs) +- [Manually Resolving Package Conflicts](#manually-resolving-package-conflicts) +- [Finalizing the Installation of `nvidia-driver-libs`](#finalizing-the-installation-of-nvidia-driver-libs) +- [Installing the CUDA Meta-Package](#installing-the-cuda-meta-package) +- [Configuring the Environment](#configuring-the-environment) +- [Verifying the Installation](#verifying-the-installation) +- [Conclusion](#conclusion) +- [Troubleshooting](#troubleshooting) +- [Additional Notes](#additional-notes) +- [References](#references) + +## Prerequisites + +- **Toolbox Installed on the Host System** `Fedora Silverblue` and `Fedora Workstation` both have toolbox by default, other distributions may need to install the [toolbox package](https://containertoolbx.org/install/). +- **NVIDIA Drivers and Graphics Card installed on Host System (optional)** To run CUDA program, such as `llama.cpp`, the host should be setup to access your NVIDIA hardware. Fedora Hosts can use the [RPM Fusion Repository](https://rpmfusion.org/Howto/NVIDIA). +- **Internet connectivity** to download packages. + +### Monitoring NVIDIA CUDA Repositories + +Before proceeding, it is advisable to check if NVIDIA has updated their CUDA repositories for your Fedora version. NVIDIA's repositories can be found at: + +- [Fedora 40 CUDA Repository](https://developer.download.nvidia.com/compute/cuda/repos/fedora40/x86_64/) +- [Fedora 41 CUDA Repository](https://developer.download.nvidia.com/compute/cuda/repos/fedora41/x86_64/) + +As of the latest update, these repositories do not contain the `cuda` meta-package or are missing essential components. + +### Using the Fedora 39 CUDA Repository + +Since the newer repositories are incomplete, we'll use the Fedora 39 repository: + +- [Fedora 39 CUDA Repository](https://developer.download.nvidia.com/compute/cuda/repos/fedora39/x86_64/) + +**Note:** Fedora 39 is no longer maintained, so we recommend using a toolbox environment to prevent system conflicts. + +## Creating a Fedora Toolbox Environment + +This guide focuses on Fedora hosts, but with small adjustments, it can work for other hosts. Using a Fedora 39 toolbox allows us to install the necessary packages without affecting the host system. + +**Note:** Toolbox is available for other systems, and even without Toolbox, it is possible to use Podman or Docker. + +We do not recommend installing on the host system, as Fedora 39 is out-of-maintenance, and instead you should upgrade to a maintained version of Fedora for your host. + +1. **Create a Fedora 39 Toolbox:** + + ```bash + toolbox create --image registry.fedoraproject.org/fedora-toolbox:39 --container fedora-toolbox-39-cuda + ``` + +2. **Enter the Toolbox:** + + ```bash + toolbox enter --container fedora-toolbox-39-cuda + ``` + + Inside the toolbox, you have root privileges and can install packages without affecting the host system. + +## Installing Essential Development Tools + +1. **Synchronize the DNF Package Manager:** + + ```bash + sudo dnf distro-sync + ``` + +2. **Install the Default Text Editor (Optional):** + + ```bash + sudo dnf install vim-default-editor --allowerasing + ``` + + The `--allowerasing` flag resolves any package conflicts. + +3. **Install Development Tools and Libraries:** + + ```bash + sudo dnf install @c-development @development-tools cmake + ``` + + This installs essential packages for compiling software, including `gcc`, `make`, and other development headers. + +## Adding the CUDA Repository + +Add the NVIDIA CUDA repository to your DNF configuration: + +```bash +sudo dnf config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/fedora39/x86_64/cuda-fedora39.repo +``` + +After adding the repository, synchronize the package manager again: + +```bash +sudo dnf distro-sync +``` + +## Installing `nvidia-driver-libs` + +Attempt to install `nvidia-driver-libs`: + +```bash +sudo dnf install nvidia-driver-libs +``` + +**Explanation:** + +- `nvidia-driver-libs` contains necessary NVIDIA driver libraries required by CUDA. +- This step might fail due to conflicts with existing NVIDIA drivers on the host system. + +## Manually Resolving Package Conflicts + +If the installation fails due to conflicts, we'll manually download and install the required packages, excluding conflicting files. + +### 1. Download the `nvidia-driver-libs` RPM + +```bash +sudo dnf download --arch x86_64 nvidia-driver-libs +``` + +You should see a file similar to: + +``` +nvidia-driver-libs-560.35.05-1.fc39.x86_64.rpm +``` + +### 2. Attempt to Install the RPM + +```bash +sudo dnf install nvidia-driver-libs-560.35.05-1.fc39.x86_64.rpm +``` + +**Expected Error:** + +Installation may fail with errors pointing to conflicts with `egl-gbm` and `egl-wayland`. + +**Note: It is important to carefully read the error messages to identify the exact paths that need to be excluded.** + +### 3. Download Dependencies + +```bash +sudo dnf download --arch x86_64 egl-gbm egl-wayland +``` + +### 4. Install `egl-gbm` with Excluded Paths + +Exclude conflicting files during installation: + +```bash +sudo rpm --install --verbose --hash \ + --excludepath=/usr/lib64/libnvidia-egl-gbm.so.1.1.2 \ + --excludepath=/usr/share/egl/egl_external_platform.d/15_nvidia_gbm.json \ + egl-gbm-1.1.2^20240919gitb24587d-3.fc39.x86_64.rpm +``` + +**Explanation:** + +- The `--excludepath` option skips installing files that conflict with existing files. +- Adjust the paths based on the error messages you receive. + +### 5. Install `egl-wayland` with Excluded Paths + +```bash +sudo rpm --install --verbose --hash \ + --excludepath=/usr/share/egl/egl_external_platform.d/10_nvidia_wayland.json \ + egl-wayland-1.1.17^20241118giteeb29e1-5.fc39.x86_64.rpm +``` + +### 6. Install `nvidia-driver-libs` with Excluded Paths + +```bash +sudo rpm --install --verbose --hash \ + --excludepath=/usr/share/glvnd/egl_vendor.d/10_nvidia.json \ + --excludepath=/usr/share/nvidia/nvoptix.bin \ + nvidia-driver-libs-560.35.05-1.fc39.x86_64.rpm +``` + +**Note:** + +- Replace the paths with the ones causing conflicts in your installation if they differ. +- The `--verbose` and `--hash` options provide detailed output during installation. + +## Finalizing the Installation of `nvidia-driver-libs` + +After manually installing the dependencies, run: + +```bash +sudo dnf install nvidia-driver-libs +``` + +You should receive a message indicating the package is already installed: + +``` +Package nvidia-driver-libs-3:560.35.05-1.fc39.x86_64 is already installed. +Dependencies resolved. +Nothing to do. +Complete! +``` + +## Installing the CUDA Meta-Package + +Now that the driver libraries are installed, proceed to install CUDA: + +```bash +sudo dnf install cuda +``` + +This installs the CUDA toolkit and associated packages. + +## Configuring the Environment + +To use CUDA, add its binary directory to your system's `PATH`. + +1. **Create a Profile Script:** + + ```bash + sudo sh -c 'echo "export PATH=\$PATH:/usr/local/cuda/bin" >> /etc/profile.d/cuda.sh' + ``` + + **Explanation:** + + - We add to `/etc/profile.d/` as the `/etc/` folder is unique to this particular container, and is not shared with other containers or the host system. + - The backslash `\` before `$PATH` ensures the variable is correctly written into the script. + +2. **Make the Script Executable:** + + ```bash + sudo chmod +x /etc/profile.d/cuda.sh + ``` + +3. **Source the Script to Update Your Environment:** + + ```bash + source /etc/profile.d/cuda.sh + ``` + + **Note:** This command updates your current shell session with the new `PATH`. The `/etc/profile.d/cuda.sh` script ensures that the CUDA binaries are available in your `PATH` for all future sessions. + +## Verifying the Installation + +To confirm that CUDA is correctly installed and configured, check the version of the NVIDIA CUDA Compiler (`nvcc`): + +```bash +nvcc --version +``` + +You should see output similar to: + +``` +nvcc: NVIDIA (R) Cuda compiler driver +Copyright (c) 2005-2024 NVIDIA Corporation +Built on Tue_Oct_29_23:50:19_PDT_2024 +Cuda compilation tools, release 12.6, V12.6.85 +Build cuda_12.6.r12.6/compiler.35059454_0 +``` + +This output confirms that the CUDA compiler is accessible and indicates the installed version. + +## Conclusion + +You have successfully set up CUDA on Fedora within a toolbox environment using the Fedora 39 CUDA repository. By manually resolving package conflicts and configuring the environment, you can develop CUDA applications without affecting your host system. + +## Troubleshooting + +- **Installation Failures:** + - If you encounter errors during installation, carefully read the error messages. They often indicate conflicting files or missing dependencies. + - Use the `--excludepath` option with `rpm` to exclude conflicting files during manual installations. + +- **Driver Conflicts:** + - Since the host system may already have NVIDIA drivers installed, conflicts can arise. Using the toolbox environment helps isolate these issues. + +- **Environment Variables Not Set:** + - If `nvcc` is not found after installation, ensure that `/usr/local/cuda/bin` is in your `PATH`. + - Run `echo $PATH` to check if the path is included. + - Re-source the profile script or open a new terminal session. + +## Additional Notes + +- **Updating CUDA in the Future:** + - Keep an eye on the official NVIDIA repositories for updates to your Fedora version. + - When an updated repository becomes available, adjust your `dnf` configuration accordingly. + +- **Building `llama.cpp`:** + - With CUDA installed, you can follow these [build instructions for `llama.cpp`](https://github.com/ggerganov/llama.cpp/blob/master/docs/build.md) to compile it with CUDA support. + - Ensure that any CUDA-specific build flags or paths are correctly set in your build configuration. + +- **Using the Toolbox Environment:** + - The toolbox environment is isolated from your host system, which helps prevent conflicts. + - Remember that system files and configurations inside the toolbox are separate from the host. By default the home directory of the user is shared between the host and the toolbox. + +--- + +**Disclaimer:** Manually installing and modifying system packages can lead to instability of the container. The above steps are provided as a guideline and may need adjustments based on your specific system configuration. Always back up important data before making significant system changes, especially as your home folder is writable and shared with he toolbox. + +**Acknowledgments:** Special thanks to the Fedora community and NVIDIA documentation for providing resources that assisted in creating this guide. + +## References + +- [Fedora Toolbox Documentation](https://docs.fedoraproject.org/en-US/fedora-silverblue/toolbox/) +- [NVIDIA CUDA Installation Guide](https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html) +- [Podman Documentation](https://podman.io/get-started) + +--- diff --git a/docs/development/HOWTO-add-model.md b/docs/development/HOWTO-add-model.md index 04c5ccbbe..8fcd70811 100644 --- a/docs/development/HOWTO-add-model.md +++ b/docs/development/HOWTO-add-model.md @@ -28,7 +28,7 @@ The required steps to implement for an HF model are: ```python @Model.register("MyModelForCausalLM") class MyModel(Model): - model_arch = gguf.MODEL_ARCH.GROK + model_arch = gguf.MODEL_ARCH.MYMODEL ``` 2. Define the layout of the GGUF tensors in [constants.py](/gguf-py/gguf/constants.py) @@ -79,14 +79,14 @@ Depending on the model configuration, tokenizer, code and tensors layout, you wi - `Model#set_vocab` - `Model#write_tensors` -NOTE: Tensor names must end with `.weight` suffix, that is the convention and several tools like `quantize` expect this to proceed the weights. +NOTE: Tensor names must end with `.weight` or `.bias` suffixes, that is the convention and several tools like `quantize` expect this to proceed the weights. ### 2. Define the model architecture in `llama.cpp` The model params and tensors layout must be defined in `llama.cpp`: 1. Define a new `llm_arch` 2. Define the tensors layout in `LLM_TENSOR_NAMES` -3. Add any non standard metadata in `llm_load_hparams` +3. Add any non-standard metadata in `llm_load_hparams` 4. Create the tensors for inference in `llm_load_tensors` 5. If the model has a RoPE operation, add the rope type in `llama_rope_type` @@ -96,9 +96,9 @@ NOTE: The dimensions in `ggml` are typically in the reverse order of the `pytorc This is the funniest part, you have to provide the inference graph implementation of the new model architecture in `llama_build_graph`. -Have a look at existing implementation like `build_llama`, `build_dbrx` or `build_bert`. +Have a look at existing implementations like `build_llama`, `build_dbrx` or `build_bert`. -When implementing a new graph, please note that the underlying `ggml` backends might not support them all, support for missing backend operations can be added in another PR. +Some `ggml` backends do not support all operations. Backend implementations can be added in a separate PR. Note: to debug the inference graph: you can use [llama-eval-callback](/examples/eval-callback/). diff --git a/docs/docker.md b/docs/docker.md index 8d90e6ded..dac9a9ec1 100644 --- a/docs/docker.md +++ b/docs/docker.md @@ -60,9 +60,9 @@ Assuming one has the [nvidia-container-toolkit](https://github.com/NVIDIA/nvidia ## Building Docker locally ```bash -docker build -t local/llama.cpp:full-cuda -f .devops/full-cuda.Dockerfile . -docker build -t local/llama.cpp:light-cuda -f .devops/llama-cli-cuda.Dockerfile . -docker build -t local/llama.cpp:server-cuda -f .devops/llama-server-cuda.Dockerfile . +docker build -t local/llama.cpp:full-cuda --target full -f .devops/cuda.Dockerfile . +docker build -t local/llama.cpp:light-cuda --target light -f .devops/cuda.Dockerfile . +docker build -t local/llama.cpp:server-cuda --target server -f .devops/cuda.Dockerfile . ``` You may want to pass in some different `ARGS`, depending on the CUDA environment supported by your container host, as well as the GPU architecture. @@ -95,9 +95,9 @@ Assuming one has the [mt-container-toolkit](https://developer.mthreads.com/musa/ ## Building Docker locally ```bash -docker build -t local/llama.cpp:full-musa -f .devops/full-musa.Dockerfile . -docker build -t local/llama.cpp:light-musa -f .devops/llama-cli-musa.Dockerfile . -docker build -t local/llama.cpp:server-musa -f .devops/llama-server-musa.Dockerfile . +docker build -t local/llama.cpp:full-musa --target full -f .devops/musa.Dockerfile . +docker build -t local/llama.cpp:light-musa --target light -f .devops/musa.Dockerfile . +docker build -t local/llama.cpp:server-musa --target server -f .devops/musa.Dockerfile . ``` You may want to pass in some different `ARGS`, depending on the MUSA environment supported by your container host, as well as the GPU architecture. diff --git a/examples/batched-bench/batched-bench.cpp b/examples/batched-bench/batched-bench.cpp index a3b21ad6b..0659ab6f1 100644 --- a/examples/batched-bench/batched-bench.cpp +++ b/examples/batched-bench/batched-bench.cpp @@ -38,7 +38,7 @@ int main(int argc, char ** argv) { llama_model_params model_params = common_model_params_to_llama(params); - llama_model * model = llama_load_model_from_file(params.model.c_str(), model_params); + llama_model * model = llama_model_load_from_file(params.model.c_str(), model_params); if (model == NULL) { fprintf(stderr , "%s: error: unable to load model\n" , __func__); @@ -50,7 +50,7 @@ int main(int argc, char ** argv) { // ensure enough sequences are available ctx_params.n_seq_max = n_pl.empty() ? 1 : *std::max_element(n_pl.begin(), n_pl.end()); - llama_context * ctx = llama_new_context_with_model(model, ctx_params); + llama_context * ctx = llama_init_from_model(model, ctx_params); if (ctx == NULL) { fprintf(stderr , "%s: error: failed to create the llama_context\n" , __func__); @@ -194,7 +194,7 @@ int main(int argc, char ** argv) { llama_batch_free(batch); llama_free(ctx); - llama_free_model(model); + llama_model_free(model); llama_backend_free(); diff --git a/examples/batched.swift/Sources/main.swift b/examples/batched.swift/Sources/main.swift index 10f2e7fd1..371917b2e 100644 --- a/examples/batched.swift/Sources/main.swift +++ b/examples/batched.swift/Sources/main.swift @@ -23,12 +23,12 @@ defer { } let model_params = llama_model_default_params() -guard let model = llama_load_model_from_file(modelPath.cString(using: .utf8), model_params) else { +guard let model = llama_model_load_from_file(modelPath.cString(using: .utf8), model_params) else { print("Failed to load model") exit(1) } defer { - llama_free_model(model) + llama_model_free(model) } var tokens = tokenize(text: prompt, add_bos: true) @@ -141,7 +141,7 @@ while n_cur <= n_len { let new_token_id = llama_sampler_sample(smpl, context, i_batch[i]) // is it an end of stream? -> mark the stream as finished - if llama_token_is_eog(model, new_token_id) || n_cur == n_len { + if llama_vocab_is_eog(model, new_token_id) || n_cur == n_len { i_batch[i] = -1 // print("") if n_parallel > 1 { diff --git a/examples/batched/batched.cpp b/examples/batched/batched.cpp index e2e01f2d5..21b95ef5e 100644 --- a/examples/batched/batched.cpp +++ b/examples/batched/batched.cpp @@ -41,17 +41,19 @@ int main(int argc, char ** argv) { llama_model_params model_params = common_model_params_to_llama(params); - llama_model * model = llama_load_model_from_file(params.model.c_str(), model_params); + llama_model * model = llama_model_load_from_file(params.model.c_str(), model_params); if (model == NULL) { LOG_ERR("%s: error: unable to load model\n" , __func__); return 1; } + const llama_vocab * vocab = llama_model_get_vocab(model); + // tokenize the prompt std::vector tokens_list; - tokens_list = common_tokenize(model, params.prompt, true); + tokens_list = common_tokenize(vocab, params.prompt, true); const int n_kv_req = tokens_list.size() + (n_predict - tokens_list.size())*n_parallel; @@ -62,7 +64,7 @@ int main(int argc, char ** argv) { ctx_params.n_ctx = n_kv_req; ctx_params.n_batch = std::max(n_predict, n_parallel); - llama_context * ctx = llama_new_context_with_model(model, ctx_params); + llama_context * ctx = llama_init_from_model(model, ctx_params); auto sparams = llama_sampler_chain_default_params(); sparams.no_perf = false; @@ -120,8 +122,8 @@ int main(int argc, char ** argv) { } llama_token decoder_start_token_id = llama_model_decoder_start_token(model); - if (decoder_start_token_id == -1) { - decoder_start_token_id = llama_token_bos(model); + if (decoder_start_token_id == LLAMA_TOKEN_NULL) { + decoder_start_token_id = llama_vocab_bos(vocab); } common_batch_clear(batch); @@ -174,7 +176,7 @@ int main(int argc, char ** argv) { const llama_token new_token_id = llama_sampler_sample(smpl, ctx, i_batch[i]); // is it an end of generation? -> mark the stream as finished - if (llama_token_is_eog(model, new_token_id) || n_cur == n_predict) { + if (llama_vocab_is_eog(vocab, new_token_id) || n_cur == n_predict) { i_batch[i] = -1; LOG("\n"); if (n_parallel > 1) { @@ -236,7 +238,7 @@ int main(int argc, char ** argv) { llama_sampler_free(smpl); llama_free(ctx); - llama_free_model(model); + llama_model_free(model); llama_backend_free(); diff --git a/examples/convert-llama2c-to-ggml/convert-llama2c-to-ggml.cpp b/examples/convert-llama2c-to-ggml/convert-llama2c-to-ggml.cpp index 736035d78..bdf0eed2a 100644 --- a/examples/convert-llama2c-to-ggml/convert-llama2c-to-ggml.cpp +++ b/examples/convert-llama2c-to-ggml/convert-llama2c-to-ggml.cpp @@ -1,4 +1,6 @@ #include "ggml.h" +#include "gguf.h" + #include "llama.h" #include "common.h" #include "log.h" @@ -689,8 +691,8 @@ static void save_as_llama_model( gguf_set_val_u32(ctx, KV_TOKENIZER_UNK_ID, UNKNOWN_TOKEN_ID); gguf_set_val_u32(ctx, KV_TOKENIZER_BOS_ID, BOS_TOKEN_ID); gguf_set_val_u32(ctx, KV_TOKENIZER_EOS_ID, EOS_TOKEN_ID); - gguf_set_val_u32(ctx, KV_TOKENIZER_SEP_ID, -1); - gguf_set_val_u32(ctx, KV_TOKENIZER_PAD_ID, -1); + gguf_set_val_u32(ctx, KV_TOKENIZER_SEP_ID, LLAMA_TOKEN_NULL); + gguf_set_val_u32(ctx, KV_TOKENIZER_PAD_ID, LLAMA_TOKEN_NULL); gguf_set_val_u32(ctx, KV_CONTEXT_LENGTH, model->hparams.n_ctx); gguf_set_val_u32(ctx, KV_EMBEDDING_LENGTH, model->hparams.n_embd); @@ -909,7 +911,7 @@ int main(int argc, char ** argv) { load_vocab(params.fn_vocab_model, &config, &vocab); struct my_llama_model model; - model.hparams.n_vocab = config.vocab_size; //llama_n_vocab(lctx); + model.hparams.n_vocab = config.vocab_size; //llama_vocab_n_vocab(lctx); model.hparams.n_ctx = params.n_ctx; model.hparams.n_embd = config.dim; //params.n_embd; model.hparams.n_ff = config.hidden_dim; diff --git a/examples/cvector-generator/cvector-generator.cpp b/examples/cvector-generator/cvector-generator.cpp index 7c9f50228..413b71d34 100644 --- a/examples/cvector-generator/cvector-generator.cpp +++ b/examples/cvector-generator/cvector-generator.cpp @@ -1,7 +1,9 @@ +#include "ggml.h" +#include "gguf.h" + #include "arg.h" #include "common.h" #include "llama.h" -#include "ggml.h" #include "pca.hpp" #include "mean.hpp" @@ -271,7 +273,9 @@ struct tokenized_prompt { size_t max_seq_len; tokenized_prompt(llama_context * ctx, std::string pos, std::string neg) { - const bool add_bos = llama_add_bos_token(llama_get_model(ctx)); + const llama_model * model = llama_get_model(ctx); + const llama_vocab * vocab = llama_model_get_vocab(model); + const bool add_bos = llama_vocab_get_add_bos(vocab); tokens_pos = common_tokenize(ctx, pos, add_bos, true); tokens_neg = common_tokenize(ctx, neg, add_bos, true); max_seq_len = std::max(tokens_pos.size(), tokens_neg.size()); @@ -419,8 +423,8 @@ int main(int argc, char ** argv) { llama_context * ctx = llama_init.context.get(); // int n_ctx = llama_n_ctx(ctx); - int n_layers = llama_n_layer(model); - int n_embd = llama_n_embd(model); + int n_layers = llama_model_n_layer(model); + int n_embd = llama_model_n_embd(model); // get model hint param (a.k.a model arch name) char model_hint[128]; diff --git a/examples/embedding/embedding.cpp b/examples/embedding/embedding.cpp index 27f75cb77..38d22c90f 100644 --- a/examples/embedding/embedding.cpp +++ b/examples/embedding/embedding.cpp @@ -105,7 +105,9 @@ int main(int argc, char ** argv) { return 1; } - const int n_ctx_train = llama_n_ctx_train(model); + const llama_vocab * vocab = llama_model_get_vocab(model); + + const int n_ctx_train = llama_model_n_ctx_train(model); const int n_ctx = llama_n_ctx(ctx); const enum llama_pooling_type pooling_type = llama_pooling_type(ctx); @@ -148,7 +150,7 @@ int main(int argc, char ** argv) { // check if the last token is SEP // it should be automatically added by the tokenizer when 'tokenizer.ggml.add_eos_token' is set to 'true' for (auto & inp : inputs) { - if (inp.empty() || inp.back() != llama_token_sep(model)) { + if (inp.empty() || inp.back() != llama_vocab_sep(vocab)) { LOG_WRN("%s: last token in the prompt is not SEP\n", __func__); LOG_WRN("%s: 'tokenizer.ggml.add_eos_token' should be set to 'true' in the GGUF header\n", __func__); } @@ -181,7 +183,7 @@ int main(int argc, char ** argv) { } // allocate output - const int n_embd = llama_n_embd(model); + const int n_embd = llama_model_n_embd(model); std::vector embeddings(n_embd_count * n_embd, 0); float * emb = embeddings.data(); diff --git a/examples/eval-callback/eval-callback.cpp b/examples/eval-callback/eval-callback.cpp index 2111c3cda..fb188f5a9 100644 --- a/examples/eval-callback/eval-callback.cpp +++ b/examples/eval-callback/eval-callback.cpp @@ -127,7 +127,10 @@ static bool ggml_debug(struct ggml_tensor * t, bool ask, void * user_data) { } static bool run(llama_context * ctx, const common_params & params) { - const bool add_bos = llama_add_bos_token(llama_get_model(ctx)); + const llama_model * model = llama_get_model(ctx); + const llama_vocab * vocab = llama_model_get_vocab(model); + + const bool add_bos = llama_vocab_get_add_bos(vocab); std::vector tokens = common_tokenize(ctx, params.prompt, add_bos); diff --git a/examples/export-lora/export-lora.cpp b/examples/export-lora/export-lora.cpp index 058b5cc86..91238e4be 100644 --- a/examples/export-lora/export-lora.cpp +++ b/examples/export-lora/export-lora.cpp @@ -1,12 +1,13 @@ -#include "arg.h" -#include "common.h" #include "ggml.h" #include "ggml-alloc.h" +#include "gguf.h" + +#include "arg.h" +#include "common.h" #include #include #include -#include #include static bool g_verbose = false; @@ -128,7 +129,7 @@ struct lora_merge_ctx { lora_merge_ctx( std::string & base_fname, - std::vector & lora_files, + std::vector & lora_files, std::string & outfile, int n_threads) : base_model(base_fname, 0), n_threads(n_threads), fout(outfile, std::ios::binary) { fout.exceptions(std::ofstream::failbit); // fail fast on write errors @@ -344,8 +345,18 @@ struct lora_merge_ctx { gf = ggml_new_graph(ctx0); struct ggml_tensor * cur = inp_base; for (size_t i = 0; i < adapters.size(); ++i) { - struct ggml_tensor * a_T = ggml_cont(ctx0, ggml_transpose(ctx0, ggml_cast(ctx0, inp_a[i], GGML_TYPE_F32))); - struct ggml_tensor * delta = ggml_mul_mat(ctx0, a_T, ggml_cast(ctx0, inp_b[i], GGML_TYPE_F32)); + struct ggml_tensor * delta; + bool is_tok_embd = string_starts_with(name_base, "token_embd"); + if (is_tok_embd) { + printf("%s : detected token embeddings tensor\n", __func__); + delta = ggml_mul_mat(ctx0, + ggml_cast(ctx0, inp_b[i], GGML_TYPE_F32), + ggml_cast(ctx0, inp_a[i], GGML_TYPE_F32)); + } else { + delta = ggml_mul_mat(ctx0, + ggml_cont(ctx0, ggml_transpose(ctx0, ggml_cast(ctx0, inp_a[i], GGML_TYPE_F32))), + ggml_cast(ctx0, inp_b[i], GGML_TYPE_F32)); + } // scale const float alpha = adapters[i]->alpha; const float rank = (float) inp_b[i]->ne[0]; diff --git a/examples/gguf-hash/gguf-hash.cpp b/examples/gguf-hash/gguf-hash.cpp index e96c75117..9523ec122 100644 --- a/examples/gguf-hash/gguf-hash.cpp +++ b/examples/gguf-hash/gguf-hash.cpp @@ -1,4 +1,5 @@ #include "ggml.h" +#include "gguf.h" #include /* abort() */ #include diff --git a/examples/gguf-split/gguf-split.cpp b/examples/gguf-split/gguf-split.cpp index 9e3d44984..ef3ceb686 100644 --- a/examples/gguf-split/gguf-split.cpp +++ b/examples/gguf-split/gguf-split.cpp @@ -1,16 +1,18 @@ +#include "ggml.h" +#include "gguf.h" #include "llama.h" #include "common.h" #include +#include +#include +#include #include +#include +#include #include #include #include -#include - -#include -#include -#include #if defined(_WIN32) #include @@ -296,7 +298,7 @@ struct split_strategy { total_size += ggml_nbytes(t); } total_size = total_size / 1000 / 1000; // convert to megabytes - printf("split %05d: n_tensors = %d, total_size = %zuM\n", i_split + 1, gguf_get_n_tensors(ctx_out), total_size); + printf("split %05d: n_tensors = %" PRIi64 ", total_size = %zuM\n", i_split + 1, gguf_get_n_tensors(ctx_out), total_size); i_split++; } } diff --git a/examples/gguf-split/tests.sh b/examples/gguf-split/tests.sh index d5a92d605..05a932227 100755 --- a/examples/gguf-split/tests.sh +++ b/examples/gguf-split/tests.sh @@ -41,7 +41,7 @@ echo PASS echo # 2b. Test the sharded model is loading properly -$MAIN --model $WORK_PATH/ggml-model-split-00001-of-00006.gguf --n-predict 32 +$MAIN -no-cnv --model $WORK_PATH/ggml-model-split-00001-of-00006.gguf --n-predict 32 echo PASS echo @@ -51,7 +51,7 @@ echo PASS echo # 3b. Test the merged model is loading properly -$MAIN --model $WORK_PATH/ggml-model-merge.gguf --n-predict 32 +$MAIN -no-cnv --model $WORK_PATH/ggml-model-merge.gguf --n-predict 32 echo PASS echo @@ -61,7 +61,7 @@ echo PASS echo # 4b. Test the sharded model is loading properly -$MAIN --model $WORK_PATH/ggml-model-split-32-tensors-00001-of-00007.gguf --n-predict 32 +$MAIN -no-cnv --model $WORK_PATH/ggml-model-split-32-tensors-00001-of-00007.gguf --n-predict 32 echo PASS echo @@ -71,7 +71,7 @@ echo #echo # 5b. Test the merged model is loading properly -#$MAIN --model $WORK_PATH/ggml-model-merge-2.gguf --n-predict 32 +#$MAIN -no-cnv --model $WORK_PATH/ggml-model-merge-2.gguf --n-predict 32 #echo PASS #echo @@ -81,7 +81,7 @@ echo PASS echo # 6b. Test the sharded model is loading properly -$MAIN --model $WORK_PATH/ggml-model-split-2G-00001-of-00002.gguf --n-predict 32 +$MAIN -no-cnv --model $WORK_PATH/ggml-model-split-2G-00001-of-00002.gguf --n-predict 32 echo PASS echo diff --git a/examples/gguf/gguf.cpp b/examples/gguf/gguf.cpp index 7498f85ef..f31989c8c 100644 --- a/examples/gguf/gguf.cpp +++ b/examples/gguf/gguf.cpp @@ -1,10 +1,9 @@ #include "ggml.h" +#include "gguf.h" #include -#include #include #include -#include #include #undef MIN @@ -135,9 +134,10 @@ static bool gguf_ex_read_0(const std::string & fname) { for (int i = 0; i < n_tensors; ++i) { const char * name = gguf_get_tensor_name (ctx, i); + const size_t size = gguf_get_tensor_size (ctx, i); const size_t offset = gguf_get_tensor_offset(ctx, i); - printf("%s: tensor[%d]: name = %s, offset = %zu\n", __func__, i, name, offset); + printf("%s: tensor[%d]: name = %s, size = %zu, offset = %zu\n", __func__, i, name, size, offset); } } @@ -182,9 +182,10 @@ static bool gguf_ex_read_1(const std::string & fname, bool check_data) { for (int i = 0; i < n_tensors; ++i) { const char * name = gguf_get_tensor_name (ctx, i); + const size_t size = gguf_get_tensor_size (ctx, i); const size_t offset = gguf_get_tensor_offset(ctx, i); - printf("%s: tensor[%d]: name = %s, offset = %zu\n", __func__, i, name, offset); + printf("%s: tensor[%d]: name = %s, size = %zu, offset = %zu\n", __func__, i, name, size, offset); } } @@ -199,7 +200,8 @@ static bool gguf_ex_read_1(const std::string & fname, bool check_data) { struct ggml_tensor * cur = ggml_get_tensor(ctx_data, name); - printf("%s: tensor[%d]: n_dims = %d, name = %s, data = %p\n", __func__, i, ggml_n_dims(cur), cur->name, cur->data); + printf("%s: tensor[%d]: n_dims = %d, ne = (%d, %d, %d, %d), name = %s, data = %p\n", + __func__, i, ggml_n_dims(cur), int(cur->ne[0]), int(cur->ne[1]), int(cur->ne[2]), int(cur->ne[3]), cur->name, cur->data); // print first 10 elements const float * data = (const float *) cur->data; @@ -215,7 +217,7 @@ static bool gguf_ex_read_1(const std::string & fname, bool check_data) { const float * data = (const float *) cur->data; for (int j = 0; j < ggml_nelements(cur); ++j) { if (data[j] != 100 + i) { - fprintf(stderr, "%s: tensor[%d]: data[%d] = %f\n", __func__, i, j, data[j]); + fprintf(stderr, "%s: tensor[%d], data[%d]: found %f, expected %f\n", __func__, i, j, data[j], float(100 + i)); gguf_free(ctx); return false; } @@ -245,6 +247,8 @@ int main(int argc, char ** argv) { check_data = false; } + srand(123456); + const std::string fname(argv[1]); const std::string mode (argv[2]); diff --git a/examples/gritlm/gritlm.cpp b/examples/gritlm/gritlm.cpp index 18a945b33..72eb46257 100644 --- a/examples/gritlm/gritlm.cpp +++ b/examples/gritlm/gritlm.cpp @@ -11,6 +11,7 @@ static std::vector> encode(llama_context * ctx, const std::ve std::vector> result; const llama_model * model = llama_get_model(ctx); + const llama_vocab * vocab = llama_model_get_vocab(model); llama_batch batch = llama_batch_init(llama_n_batch(ctx), 0, 1); @@ -19,16 +20,16 @@ static std::vector> encode(llama_context * ctx, const std::ve const std::string input_string = instruction + sentences[i]; - std::vector inputs = common_tokenize(model, input_string, true, false); + std::vector inputs = common_tokenize(vocab, input_string, true, false); const int32_t n_toks = inputs.size(); // GritLM seems to have EOS = "" // https://github.com/ContextualAI/gritlm/blob/92025b16534712b31b3c4aaaf069350e222bd5f8/gritlm/gritlm.py#L18 - // inputs.push_back(llama_token_eos(model)); + // inputs.push_back(llama_vocab_eos(vocab)); // we want to ignore instruction tokens for mean pooling - const int32_t n_inst = common_tokenize(model, instruction, true, false).size(); + const int32_t n_inst = common_tokenize(vocab, instruction, true, false).size(); #ifdef GRIT_DEBUG // debug tokens - should be matching as referenced in the GritLM sample @@ -52,7 +53,7 @@ static std::vector> encode(llama_context * ctx, const std::ve llama_decode(ctx, batch); // get embedding dimensions - uint64_t n_embd = llama_n_embd(model); + uint64_t n_embd = llama_model_n_embd(model); // allocate embedding output std::vector emb_unorm(n_embd, 0.0f); @@ -97,7 +98,9 @@ static std::string generate(llama_context * ctx, llama_sampler * smpl, const std std::string result; const llama_model * model = llama_get_model(ctx); - llama_token eos_token = llama_token_eos(model); + const llama_vocab * vocab = llama_model_get_vocab(model); + + llama_token eos_token = llama_vocab_eos(vocab); llama_kv_cache_clear(ctx); llama_set_embeddings(ctx, false); @@ -105,7 +108,7 @@ static std::string generate(llama_context * ctx, llama_sampler * smpl, const std llama_batch bat = llama_batch_init(llama_n_batch(ctx), 0, 1); - std::vector inputs = common_tokenize(model, prompt, false, true); + std::vector inputs = common_tokenize(vocab, prompt, false, true); int32_t i_current_token = 0; while (true) { @@ -165,10 +168,10 @@ int main(int argc, char * argv[]) { llama_backend_init(); - llama_model * model = llama_load_model_from_file(params.model.c_str(), mparams); + llama_model * model = llama_model_load_from_file(params.model.c_str(), mparams); // create generation context - llama_context * ctx = llama_new_context_with_model(model, cparams); + llama_context * ctx = llama_init_from_model(model, cparams); auto sparams = llama_sampler_chain_default_params(); @@ -197,7 +200,7 @@ int main(int argc, char * argv[]) { const std::vector> d_rep = encode(ctx, documents, gritlm_instruction("")); const std::vector> q_rep = encode(ctx, queries, gritlm_instruction(instruction)); - const int n_embd = llama_n_embd(model); + const int n_embd = llama_model_n_embd(model); const float cosine_sim_q0_d0 = common_embd_similarity_cos(q_rep[0].data(), d_rep[0].data(), n_embd); const float cosine_sim_q0_d1 = common_embd_similarity_cos(q_rep[0].data(), d_rep[1].data(), n_embd); @@ -219,7 +222,7 @@ int main(int argc, char * argv[]) { llama_sampler_free(smpl); llama_free(ctx); - llama_free_model(model); + llama_model_free(model); llama_backend_free(); return 0; diff --git a/examples/imatrix/imatrix.cpp b/examples/imatrix/imatrix.cpp index 588114ecd..b5f3feb9f 100644 --- a/examples/imatrix/imatrix.cpp +++ b/examples/imatrix/imatrix.cpp @@ -7,7 +7,6 @@ #include #include #include -#include #include #include #include @@ -40,7 +39,7 @@ public: void set_params(common_params params) { m_params = std::move(params); } bool collect_imatrix(struct ggml_tensor * t, bool ask, void * user_data); void save_imatrix(int ncall = -1) const; - bool load_imatrix(const char * file_name); + bool load_imatrix(const char * fname); private: std::unordered_map m_stats; common_params m_params; @@ -429,10 +428,13 @@ static void process_logits( } static bool compute_imatrix(llama_context * ctx, const common_params & params) { - const bool add_bos = llama_add_bos_token(llama_get_model(ctx)); + const llama_model * model = llama_get_model(ctx); + const llama_vocab * vocab = llama_model_get_vocab(model); + + const bool add_bos = llama_vocab_get_add_bos(vocab); const int n_ctx = llama_n_ctx(ctx); - GGML_ASSERT(!llama_add_eos_token(llama_get_model(ctx))); + GGML_ASSERT(!llama_vocab_get_add_eos(vocab)); auto tim1 = std::chrono::high_resolution_clock::now(); LOG_INF("%s: tokenizing the input ..\n", __func__); @@ -468,7 +470,7 @@ static bool compute_imatrix(llama_context * ctx, const common_params & params) { const int n_chunk_max = tokens.size() / n_ctx; const int n_chunk = params.n_chunks < 0 ? n_chunk_max : std::min(params.n_chunks, n_chunk_max); - const int n_vocab = llama_n_vocab(llama_get_model(ctx)); + const int n_vocab = llama_vocab_n_tokens(vocab); const int n_batch = params.n_batch; int count = 0; @@ -508,7 +510,7 @@ static bool compute_imatrix(llama_context * ctx, const common_params & params) { // add BOS token for the first batch of each chunk if (add_bos && j == 0) { - tokens[batch_start] = llama_token_bos(llama_get_model(ctx)); + tokens[batch_start] = llama_vocab_bos(vocab); } common_batch_clear(batch); @@ -627,7 +629,7 @@ int main(int argc, char ** argv) { return 1; } - const int n_ctx_train = llama_n_ctx_train(model); + const int n_ctx_train = llama_model_n_ctx_train(model); if (params.n_ctx > n_ctx_train) { LOG_WRN("%s: model was trained on only %d context tokens (%d specified)\n", __func__, n_ctx_train, params.n_ctx); diff --git a/examples/infill/infill.cpp b/examples/infill/infill.cpp index d460be314..489a208b6 100644 --- a/examples/infill/infill.cpp +++ b/examples/infill/infill.cpp @@ -139,7 +139,9 @@ int main(int argc, char ** argv) { return 1; } - const int n_ctx_train = llama_n_ctx_train(model); + const llama_vocab * vocab = llama_model_get_vocab(model); + + const int n_ctx_train = llama_model_n_ctx_train(model); const int n_ctx = llama_n_ctx(ctx); LOG_DBG("n_ctx: %d\n", n_ctx); @@ -152,28 +154,28 @@ int main(int argc, char ** argv) { LOG_INF("\n"); LOG_INF("%s\n", common_params_get_system_info(params).c_str()); } - const bool add_bos = llama_add_bos_token(model); - GGML_ASSERT(!llama_add_eos_token(model)); + const bool add_bos = llama_vocab_get_add_bos(vocab); + GGML_ASSERT(!llama_vocab_get_add_eos(vocab)); std::vector embd_inp; std::vector embd_end; std::vector inp_pfx = common_tokenize(ctx, params.input_prefix, false); std::vector inp_sfx = common_tokenize(ctx, params.input_suffix, false); - GGML_ASSERT(llama_token_fim_pre(model) >= 0); - GGML_ASSERT(llama_token_fim_suf(model) >= 0); + GGML_ASSERT(llama_vocab_fim_pre(vocab) >= 0); + GGML_ASSERT(llama_vocab_fim_suf(vocab) >= 0); - inp_pfx.insert(inp_pfx.begin(), llama_token_fim_pre(model)); - inp_sfx.insert(inp_sfx.begin(), llama_token_fim_suf(model)); + inp_pfx.insert(inp_pfx.begin(), llama_vocab_fim_pre(vocab)); + inp_sfx.insert(inp_sfx.begin(), llama_vocab_fim_suf(vocab)); embd_inp = params.spm_infill ? inp_sfx : inp_pfx; embd_end = params.spm_infill ? inp_pfx : inp_sfx; if (add_bos) { - embd_inp.insert(embd_inp.begin(), llama_token_bos(model)); + embd_inp.insert(embd_inp.begin(), llama_vocab_bos(vocab)); } embd_inp.insert(embd_inp.end(), embd_end.begin(), embd_end.end()); - const llama_token middle_token = llama_token_fim_mid(model); + const llama_token middle_token = llama_vocab_fim_mid(vocab); if (middle_token >= 0) { embd_inp.push_back(middle_token); } @@ -185,7 +187,7 @@ int main(int argc, char ** argv) { // Should not run without any tokens if (embd_inp.empty()) { - embd_inp.push_back(llama_token_bos(model)); + embd_inp.push_back(llama_vocab_bos(vocab)); LOG_WRN("embd_inp was considered empty and bos was added: %s\n", string_from(ctx, embd_inp).c_str()); } @@ -420,10 +422,10 @@ int main(int argc, char ** argv) { // if not currently processing queued inputs; if ((int) embd_inp.size() <= n_consumed) { // deal with eot token in infill mode - if ((common_sampler_last(smpl) == llama_token_eot(model) || is_interacting) && params.interactive){ + if ((common_sampler_last(smpl) == llama_vocab_eot(vocab) || is_interacting) && params.interactive){ if (is_interacting && !params.interactive_first) { // print an eot token - LOG("%s", common_token_to_piece(ctx, llama_token_eot(model)).c_str()); + LOG("%s", common_token_to_piece(ctx, llama_vocab_eot(vocab)).c_str()); } LOG("\n"); console::set_display(console::user_input); @@ -463,13 +465,13 @@ int main(int argc, char ** argv) { std::vector inp_pfx = common_tokenize(ctx, params.input_prefix, false); std::vector inp_sfx = common_tokenize(ctx, params.input_suffix, false); - inp_pfx.insert(inp_pfx.begin(), llama_token_fim_pre(model)); - inp_sfx.insert(inp_sfx.begin(), llama_token_fim_suf(model)); + inp_pfx.insert(inp_pfx.begin(), llama_vocab_fim_pre(vocab)); + inp_sfx.insert(inp_sfx.begin(), llama_vocab_fim_suf(vocab)); embd_inp = params.spm_infill ? inp_sfx : inp_pfx; embd_end = params.spm_infill ? inp_pfx : inp_sfx; if (add_bos) { - embd_inp.insert(embd_inp.begin(), llama_token_bos(model)); + embd_inp.insert(embd_inp.begin(), llama_vocab_bos(vocab)); } embd_inp.insert(embd_inp.end(), embd_end.begin(), embd_end.end()); @@ -484,7 +486,7 @@ int main(int argc, char ** argv) { is_interacting = false; } // deal with end of generation tokens in interactive mode - else if (llama_token_is_eog(model, common_sampler_last(smpl))) { + else if (llama_vocab_is_eog(vocab, common_sampler_last(smpl))) { LOG_DBG("found EOS token\n"); if (params.interactive) { @@ -500,7 +502,7 @@ int main(int argc, char ** argv) { if (params.input_prefix_bos) { LOG_DBG("adding input prefix BOS token\n"); - embd_inp.push_back(llama_token_bos(model)); + embd_inp.push_back(llama_vocab_bos(vocab)); } std::string buffer; @@ -563,7 +565,7 @@ int main(int argc, char ** argv) { } // end of generation - if (!embd.empty() && llama_token_is_eog(model, embd.back()) && !params.interactive) { + if (!embd.empty() && llama_vocab_is_eog(vocab, embd.back()) && !params.interactive) { break; } @@ -575,7 +577,7 @@ int main(int argc, char ** argv) { } } if (!params.interactive && n_remain <= 0) { - LOG("%s", common_token_to_piece(ctx, llama_token_eot(model)).c_str()); + LOG("%s", common_token_to_piece(ctx, llama_vocab_eot(vocab)).c_str()); } LOG("\n"); diff --git a/examples/llama-bench/llama-bench.cpp b/examples/llama-bench/llama-bench.cpp index 2338ad106..4ac19ca86 100644 --- a/examples/llama-bench/llama-bench.cpp +++ b/examples/llama-bench/llama-bench.cpp @@ -683,7 +683,7 @@ struct cmd_params_instance { bool cpu_strict; int poll; int n_gpu_layers; - std::string rpc_servers; + std::string rpc_servers_str; llama_split_mode split_mode; int main_gpu; bool no_kv_offload; @@ -696,8 +696,37 @@ struct cmd_params_instance { llama_model_params mparams = llama_model_default_params(); mparams.n_gpu_layers = n_gpu_layers; - if (!rpc_servers.empty()) { - mparams.rpc_servers = rpc_servers.c_str(); + if (!rpc_servers_str.empty()) { + auto rpc_servers = string_split(rpc_servers_str, ','); + + // add RPC devices + if (!rpc_servers.empty()) { + ggml_backend_reg_t rpc_reg = ggml_backend_reg_by_name("RPC"); + if (!rpc_reg) { + fprintf(stderr, "%s: failed to find RPC backend\n", __func__); + exit(1); + } + + typedef ggml_backend_dev_t (*ggml_backend_rpc_add_device_t)(const char * endpoint); + ggml_backend_rpc_add_device_t ggml_backend_rpc_add_device_fn = (ggml_backend_rpc_add_device_t) ggml_backend_reg_get_proc_address(rpc_reg, "ggml_backend_rpc_add_device"); + if (!ggml_backend_rpc_add_device_fn) { + fprintf(stderr, "%s: failed to find RPC device add function\n", __func__); + exit(1); + } + static std::vector devices; + devices.clear(); + for (const std::string & server : rpc_servers) { + ggml_backend_dev_t dev = ggml_backend_rpc_add_device_fn(server.c_str()); + if (dev) { + devices.push_back(dev); + } else { + fprintf(stderr, "%s: failed to add RPC device for server '%s'\n", __func__, server.c_str()); + exit(1); + } + } + devices.push_back(nullptr); + mparams.devices = devices.data(); + } } mparams.split_mode = split_mode; mparams.main_gpu = main_gpu; @@ -708,7 +737,7 @@ struct cmd_params_instance { } bool equal_mparams(const cmd_params_instance & other) const { - return model == other.model && n_gpu_layers == other.n_gpu_layers && rpc_servers == other.rpc_servers && + return model == other.model && n_gpu_layers == other.n_gpu_layers && rpc_servers_str == other.rpc_servers_str && split_mode == other.split_mode && main_gpu == other.main_gpu && use_mmap == other.use_mmap && tensor_split == other.tensor_split; } @@ -1401,7 +1430,8 @@ static void test_prompt(llama_context * ctx, int n_prompt, int n_batch, int n_th llama_set_n_threads(ctx, n_threads, n_threads); const llama_model * model = llama_get_model(ctx); - const int32_t n_vocab = llama_n_vocab(model); + const llama_vocab * vocab = llama_model_get_vocab(model); + const int32_t n_vocab = llama_vocab_n_tokens(vocab); std::vector tokens(n_batch); @@ -1409,7 +1439,7 @@ static void test_prompt(llama_context * ctx, int n_prompt, int n_batch, int n_th while (n_processed < n_prompt) { int n_tokens = std::min(n_prompt - n_processed, n_batch); - tokens[0] = n_processed == 0 && llama_add_bos_token(model) ? llama_token_bos(model) : std::rand() % n_vocab; + tokens[0] = n_processed == 0 && llama_vocab_get_add_bos(vocab) ? llama_vocab_bos(vocab) : std::rand() % n_vocab; for (int i = 1; i < n_tokens; i++) { tokens[i] = std::rand() % n_vocab; } @@ -1424,9 +1454,10 @@ static void test_gen(llama_context * ctx, int n_gen, int n_threads) { llama_set_n_threads(ctx, n_threads, n_threads); const llama_model * model = llama_get_model(ctx); - const int32_t n_vocab = llama_n_vocab(model); + const llama_vocab * vocab = llama_model_get_vocab(model); + const int32_t n_vocab = llama_vocab_n_tokens(vocab); - llama_token token = llama_add_bos_token(model) ? llama_token_bos(model) : std::rand() % n_vocab; + llama_token token = llama_vocab_get_add_bos(vocab) ? llama_vocab_bos(vocab) : std::rand() % n_vocab; for (int i = 0; i < n_gen; i++) { llama_decode(ctx, llama_batch_get_one(&token, 1)); @@ -1526,10 +1557,10 @@ int main(int argc, char ** argv) { // keep the same model between tests when possible if (!lmodel || !prev_inst || !inst.equal_mparams(*prev_inst)) { if (lmodel) { - llama_free_model(lmodel); + llama_model_free(lmodel); } - lmodel = llama_load_model_from_file(inst.model.c_str(), inst.to_llama_mparams()); + lmodel = llama_model_load_from_file(inst.model.c_str(), inst.to_llama_mparams()); if (lmodel == NULL) { fprintf(stderr, "%s: error: failed to load model '%s'\n", __func__, inst.model.c_str()); return 1; @@ -1537,10 +1568,10 @@ int main(int argc, char ** argv) { prev_inst = &inst; } - llama_context * ctx = llama_new_context_with_model(lmodel, inst.to_llama_cparams()); + llama_context * ctx = llama_init_from_model(lmodel, inst.to_llama_cparams()); if (ctx == NULL) { fprintf(stderr, "%s: error: failed to create context with model '%s'\n", __func__, inst.model.c_str()); - llama_free_model(lmodel); + llama_model_free(lmodel); return 1; } @@ -1626,7 +1657,7 @@ int main(int argc, char ** argv) { ggml_threadpool_free_fn(threadpool); } - llama_free_model(lmodel); + llama_model_free(lmodel); if (p) { p->print_footer(); diff --git a/examples/llama.android/llama/src/main/cpp/llama-android.cpp b/examples/llama.android/llama/src/main/cpp/llama-android.cpp index 66ec2aeeb..2a73983a9 100644 --- a/examples/llama.android/llama/src/main/cpp/llama-android.cpp +++ b/examples/llama.android/llama/src/main/cpp/llama-android.cpp @@ -87,7 +87,7 @@ Java_android_llama_cpp_LLamaAndroid_load_1model(JNIEnv *env, jobject, jstring fi auto path_to_model = env->GetStringUTFChars(filename, 0); LOGi("Loading model from %s", path_to_model); - auto model = llama_load_model_from_file(path_to_model, model_params); + auto model = llama_model_load_from_file(path_to_model, model_params); env->ReleaseStringUTFChars(filename, path_to_model); if (!model) { @@ -102,7 +102,7 @@ Java_android_llama_cpp_LLamaAndroid_load_1model(JNIEnv *env, jobject, jstring fi extern "C" JNIEXPORT void JNICALL Java_android_llama_cpp_LLamaAndroid_free_1model(JNIEnv *, jobject, jlong model) { - llama_free_model(reinterpret_cast(model)); + llama_model_free(reinterpret_cast(model)); } extern "C" @@ -347,6 +347,7 @@ Java_android_llama_cpp_LLamaAndroid_completion_1init( jlong context_pointer, jlong batch_pointer, jstring jtext, + jboolean format_chat, jint n_len ) { @@ -356,7 +357,8 @@ Java_android_llama_cpp_LLamaAndroid_completion_1init( const auto context = reinterpret_cast(context_pointer); const auto batch = reinterpret_cast(batch_pointer); - const auto tokens_list = common_tokenize(context, text, 1); + bool parse_special = (format_chat == JNI_TRUE); + const auto tokens_list = common_tokenize(context, text, true, parse_special); auto n_ctx = llama_n_ctx(context); auto n_kv_req = tokens_list.size() + (n_len - tokens_list.size()); @@ -368,7 +370,7 @@ Java_android_llama_cpp_LLamaAndroid_completion_1init( } for (auto id : tokens_list) { - LOGi("%s", common_token_to_piece(context, id).c_str()); + LOGi("token: `%s`-> %d ", common_token_to_piece(context, id).c_str(), id); } common_batch_clear(*batch); @@ -405,6 +407,7 @@ Java_android_llama_cpp_LLamaAndroid_completion_1loop( const auto batch = reinterpret_cast(batch_pointer); const auto sampler = reinterpret_cast(sampler_pointer); const auto model = llama_get_model(context); + const auto vocab = llama_model_get_vocab(model); if (!la_int_var) la_int_var = env->GetObjectClass(intvar_ncur); if (!la_int_var_value) la_int_var_value = env->GetMethodID(la_int_var, "getValue", "()I"); @@ -414,7 +417,7 @@ Java_android_llama_cpp_LLamaAndroid_completion_1loop( const auto new_token_id = llama_sampler_sample(sampler, context, -1); const auto n_cur = env->CallIntMethod(intvar_ncur, la_int_var_value); - if (llama_token_is_eog(model, new_token_id) || n_cur == n_len) { + if (llama_vocab_is_eog(vocab, new_token_id) || n_cur == n_len) { return nullptr; } diff --git a/examples/llama.android/llama/src/main/java/android/llama/cpp/LLamaAndroid.kt b/examples/llama.android/llama/src/main/java/android/llama/cpp/LLamaAndroid.kt index cf520e459..b964d93e3 100644 --- a/examples/llama.android/llama/src/main/java/android/llama/cpp/LLamaAndroid.kt +++ b/examples/llama.android/llama/src/main/java/android/llama/cpp/LLamaAndroid.kt @@ -65,6 +65,7 @@ class LLamaAndroid { context: Long, batch: Long, text: String, + formatChat: Boolean, nLen: Int ): Int @@ -115,10 +116,10 @@ class LLamaAndroid { } } - fun send(message: String): Flow = flow { + fun send(message: String, formatChat: Boolean = false): Flow = flow { when (val state = threadLocalState.get()) { is State.Loaded -> { - val ncur = IntVar(completion_init(state.context, state.batch, message, nlen)) + val ncur = IntVar(completion_init(state.context, state.batch, message, formatChat, nlen)) while (ncur.value <= nlen) { val str = completion_loop(state.context, state.batch, state.sampler, nlen, ncur) if (str == null) { diff --git a/examples/llama.swiftui/llama.cpp.swift/LibLlama.swift b/examples/llama.swiftui/llama.cpp.swift/LibLlama.swift index 998c673d5..477c3e6f2 100644 --- a/examples/llama.swiftui/llama.cpp.swift/LibLlama.swift +++ b/examples/llama.swiftui/llama.cpp.swift/LibLlama.swift @@ -52,8 +52,8 @@ actor LlamaContext { deinit { llama_sampler_free(sampling) llama_batch_free(batch) + llama_model_free(model) llama_free(context) - llama_free_model(model) llama_backend_free() } @@ -65,7 +65,7 @@ actor LlamaContext { model_params.n_gpu_layers = 0 print("Running on simulator, force use n_gpu_layers = 0") #endif - let model = llama_load_model_from_file(path, model_params) + let model = llama_model_load_from_file(path, model_params) guard let model else { print("Could not load model at \(path)") throw LlamaError.couldNotInitializeContext @@ -151,7 +151,7 @@ actor LlamaContext { new_token_id = llama_sampler_sample(sampling, context, batch.n_tokens - 1) - if llama_token_is_eog(model, new_token_id) || n_cur == n_len { + if llama_vocab_is_eog(model, new_token_id) || n_cur == n_len { print("\n") is_done = true let new_token_str = String(cString: temporary_invalid_cchars + [0]) diff --git a/examples/llava/README-minicpmo2.6.md b/examples/llava/README-minicpmo2.6.md new file mode 100644 index 000000000..8713a43d6 --- /dev/null +++ b/examples/llava/README-minicpmo2.6.md @@ -0,0 +1,46 @@ +## MiniCPM-o 2.6 +Currently, this readme only supports minicpm-omni's image capabilities, and we will update the full-mode support as soon as possible. + +### Prepare models and code + +Download [MiniCPM-o-2_6](https://huggingface.co/openbmb/MiniCPM-o-2_6) PyTorch model from huggingface to "MiniCPM-o-2_6" folder. + +Clone llama.cpp: +```bash +git clone git@github.com:OpenBMB/llama.cpp.git +cd llama.cpp +git checkout minicpm-omni +``` + +### Usage of MiniCPM-o 2.6 + +Convert PyTorch model to gguf files (You can also download the converted [gguf](https://huggingface.co/openbmb/MiniCPM-o-2_6-gguf) by us) + +```bash +python ./examples/llava/minicpmv-surgery.py -m ../MiniCPM-o-2_6 +python ./examples/llava/minicpmv-convert-image-encoder-to-gguf.py -m ../MiniCPM-o-2_6 --minicpmv-projector ../MiniCPM-o-2_6/minicpmv.projector --output-dir ../MiniCPM-o-2_6/ --image-mean 0.5 0.5 0.5 --image-std 0.5 0.5 0.5 --minicpmv_version 4 +python ./convert_hf_to_gguf.py ../MiniCPM-o-2_6/model + +# quantize int4 version +./llama-quantize ../MiniCPM-o-2_6/model/ggml-model-f16.gguf ../MiniCPM-o-2_6/model/ggml-model-Q4_K_M.gguf Q4_K_M +``` + +Build llama.cpp using `CMake`: +https://github.com/ggerganov/llama.cpp/blob/master/docs/build.md + +```bash +cmake -B build +cmake --build build --config Release +``` + +Inference on Linux or Mac +``` +# run f16 version +./llama-minicpmv-cli -m ../MiniCPM-o-2_6/model/ggml-model-f16.gguf --mmproj ../MiniCPM-o-2_6/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg -p "What is in the image?" + +# run quantized int4 version +./llama-minicpmv-cli -m ../MiniCPM-o-2_6/model/ggml-model-Q4_K_M.gguf --mmproj ../MiniCPM-o-2_6/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg -p "What is in the image?" + +# or run in interactive mode +./llama-minicpmv-cli -m ../MiniCPM-o-2_6/model/ggml-model-Q4_K_M.gguf --mmproj ../MiniCPM-o-2_6/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg -i +``` diff --git a/examples/llava/clip.cpp b/examples/llava/clip.cpp index 3cd0d2fa8..24073c5a9 100644 --- a/examples/llava/clip.cpp +++ b/examples/llava/clip.cpp @@ -7,6 +7,7 @@ #include "ggml-cpu.h" #include "ggml-alloc.h" #include "ggml-backend.h" +#include "gguf.h" //#ifdef GGML_USE_CUDA //#include "ggml-cuda.h" @@ -262,7 +263,7 @@ static std::string gguf_kv_to_str(const struct gguf_context * ctx_gguf, int i) { { const enum gguf_type arr_type = gguf_get_arr_type(ctx_gguf, i); int arr_n = gguf_get_arr_n(ctx_gguf, i); - const void * data = gguf_get_arr_data(ctx_gguf, i); + const void * data = arr_type == GGUF_TYPE_STRING ? nullptr : gguf_get_arr_data(ctx_gguf, i); std::stringstream ss; ss << "["; for (int j = 0; j < arr_n; j++) { @@ -717,6 +718,9 @@ static ggml_cgraph * clip_image_build_graph(clip_ctx * ctx, const clip_image_f32 else if (ctx->minicpmv_version == 3) { pos_embed = ggml_new_tensor_3d(ctx0, GGML_TYPE_F32, 3584, pos_w * pos_h, 1); } + else if (ctx->minicpmv_version == 4) { + pos_embed = ggml_new_tensor_3d(ctx0, GGML_TYPE_F32, 3584, pos_w * pos_h, 1); + } ggml_set_name(pos_embed, "pos_embed"); ggml_set_input(pos_embed); } @@ -1052,6 +1056,11 @@ static ggml_cgraph * clip_image_build_graph(clip_ctx * ctx, const clip_image_f32 n_head = hidden_size/d_head; num_query = 64; } + else if (ctx->minicpmv_version == 4) { + hidden_size = 3584; + n_head = hidden_size/d_head; + num_query = 64; + } struct ggml_tensor * Q = ggml_add(ctx0, ggml_mul_mat(ctx0, model.mm_model_attn_q_w, q), model.mm_model_attn_q_b); Q = ggml_scale_inplace(ctx0, Q, 1.0f / sqrt((float)d_head)); @@ -2040,6 +2049,7 @@ static std::vector> uhd_slice_image(const clip_imag images[images.size()-1].push_back(patch); } } + clip_image_u8_free(refine_image); } return images; } @@ -2078,6 +2088,13 @@ bool clip_image_preprocess(struct clip_ctx * ctx, const clip_image_u8 * img, cli clip_image_f32_free(res); } } + for (size_t i = 0; i < imgs.size(); ++i) { + for (size_t j = 0; j < imgs[i].size(); ++j) { + if (imgs[i][j] != nullptr) { + clip_image_u8_free(imgs[i][j]); + } + } + } return true; } else if (ctx->has_qwen2vl_merger) { @@ -2334,6 +2351,9 @@ int clip_n_patches_by_img(const struct clip_ctx * ctx, struct clip_image_f32 * i else if (ctx->minicpmv_version == 3) { n_patches = 64; } + else if (ctx->minicpmv_version == 4) { + n_patches = 64; + } } else if (ctx->proj_type == PROJECTOR_TYPE_MERGER) { int patch_size = params.patch_size * 2; int x_patch = img->nx / patch_size + (int)(img->nx % patch_size > 0); @@ -2513,8 +2533,8 @@ bool clip_image_batch_encode(clip_ctx * ctx, const int n_threads, const clip_ima // -> https://huggingface.co/HuggingFaceM4/siglip-so400m-14-980-flash-attn2-navit/blob/d66538faeba44480d0bfaa42145eef26f9423199/modeling_siglip.py#L316 struct ggml_tensor * positions = ggml_graph_get_tensor(gf, "positions"); int* positions_data = (int*)malloc(ggml_nbytes(positions)); - int bucket_coords_h[70]; - int bucket_coords_w[70]; + int bucket_coords_h[1024]; + int bucket_coords_w[1024]; for (int i = 0; i < pos_h; i++){ bucket_coords_h[i] = std::floor(70.0*i/pos_h); } @@ -2542,6 +2562,9 @@ bool clip_image_batch_encode(clip_ctx * ctx, const int n_threads, const clip_ima else if (ctx->minicpmv_version == 3) { embed_dim = 3584; } + else if (ctx->minicpmv_version == 4) { + embed_dim = 3584; + } auto pos_embed_t = get_2d_sincos_pos_embed(embed_dim, std::make_pair(pos_w, pos_h)); float * pos_embed_data = (float *)malloc(ggml_nbytes(pos_embed)); @@ -2734,7 +2757,8 @@ bool clip_model_quantize(const char * fname_inp, const char * fname_out, const i total_size_org += orig_size; total_size_new += new_size; gguf_set_tensor_type(ctx_out, name.c_str(), new_type); - gguf_set_tensor_data(ctx_out, name.c_str(), new_data, new_size); + GGML_ASSERT(gguf_get_tensor_size(ctx_out, gguf_find_tensor(ctx_out, name.c_str())) == new_size); + gguf_set_tensor_data(ctx_out, name.c_str(), new_data); fout.write((const char *)new_data, new_size); size_t pad = GGML_PAD(new_size, gguf_get_alignment(ctx_out)) - new_size; for (size_t j = 0; j < pad; ++j) { @@ -2784,6 +2808,9 @@ int clip_n_mmproj_embd(const struct clip_ctx * ctx) { else if (ctx->minicpmv_version == 3) { return 3584; } + else if (ctx->minicpmv_version == 4) { + return 3584; + } } if (ctx->proj_type == PROJECTOR_TYPE_MERGER) { return ctx->vision_model.mm_1_b->ne[0]; diff --git a/examples/llava/llava-cli.cpp b/examples/llava/llava-cli.cpp index 2691c6e6b..40aa0876f 100644 --- a/examples/llava/llava-cli.cpp +++ b/examples/llava/llava-cli.cpp @@ -47,8 +47,12 @@ static const char * sample(struct common_sampler * smpl, int * n_past) { const llama_token id = common_sampler_sample(smpl, ctx_llama, -1); common_sampler_accept(smpl, id, true); + + const llama_model * model = llama_get_model(ctx_llama); + const llama_vocab * vocab = llama_model_get_vocab(model); + static std::string ret; - if (llama_token_is_eog(llama_get_model(ctx_llama), id)) { + if (llama_vocab_is_eog(vocab, id)) { ret = ""; } else { ret = common_token_to_piece(ctx_llama, id); @@ -221,7 +225,7 @@ static struct llama_model * llava_init(common_params * params) { llama_model_params model_params = common_model_params_to_llama(*params); - llama_model * model = llama_load_model_from_file(params->model.c_str(), model_params); + llama_model * model = llama_model_load_from_file(params->model.c_str(), model_params); if (model == NULL) { LOG_ERR("%s: unable to load model\n" , __func__); return NULL; @@ -239,11 +243,10 @@ static struct llava_context * llava_init_context(common_params * params, llama_m auto ctx_clip = clip_model_load(clip_path, /*verbosity=*/ 1); - llama_context_params ctx_params = common_context_params_to_llama(*params); ctx_params.n_ctx = params->n_ctx < 2048 ? 2048 : params->n_ctx; // we need a longer context size to process image embeddings - llama_context * ctx_llama = llama_new_context_with_model(model, ctx_params); + llama_context * ctx_llama = llama_init_from_model(model, ctx_params); if (ctx_llama == NULL) { LOG_ERR("%s: failed to create the llama_context\n" , __func__); @@ -265,7 +268,7 @@ static void llava_free(struct llava_context * ctx_llava) { } llama_free(ctx_llava->ctx_llama); - llama_free_model(ctx_llava->model); + llama_model_free(ctx_llava->model); llama_backend_free(); } @@ -323,7 +326,7 @@ int main(int argc, char ** argv) { } } - llama_free_model(model); + llama_model_free(model); return 0; } diff --git a/examples/llava/llava.cpp b/examples/llava/llava.cpp index 16f30c56c..2cac7933d 100644 --- a/examples/llava/llava.cpp +++ b/examples/llava/llava.cpp @@ -216,7 +216,7 @@ static bool clip_llava_handle_patches(clip_ctx * ctx_clip, std::vector return true; } -static clip_image_f32 * only_v2_5_reshape_by_patch(clip_image_f32 * image, int patch_size) { +static clip_image_f32 * reshape_by_patch(clip_image_f32 * image, int patch_size) { int width = image->nx; int height = image->ny; int num_patches = (height / patch_size) * (width / patch_size); @@ -277,13 +277,7 @@ static bool encode_image_with_clip(clip_ctx * ctx_clip, int n_threads, const cli encoded = clip_image_encode(ctx_clip, n_threads, &img_res_v.data[i], image_embd_v[i]); } else { - int has_minicpmv_projector = clip_is_minicpmv(ctx_clip); - if (has_minicpmv_projector == 2) { - encoded = clip_image_encode(ctx_clip, n_threads, only_v2_5_reshape_by_patch(&img_res_v.data[i], patch_size), image_embd_v[i]); - } - else if (has_minicpmv_projector == 3) { - encoded = clip_image_encode(ctx_clip, n_threads, &img_res_v.data[i], image_embd_v[i]); - } + encoded = clip_image_encode(ctx_clip, n_threads, reshape_by_patch(&img_res_v.data[i], patch_size), image_embd_v[i]); } if (!encoded) { @@ -313,6 +307,9 @@ static bool encode_image_with_clip(clip_ctx * ctx_clip, int n_threads, const cli load_image_size->height = img->ny; clip_add_load_image_size(ctx_clip, load_image_size); LOG_INF("%s: load_image_size %d %d\n", __func__, load_image_size->width, load_image_size->height); + delete[] img_res_v.data; + img_res_v.size = 0; + img_res_v.data = nullptr; } else if (strcmp(mm_patch_merge_type, "spatial_unpad") != 0) { // flat / default llava-1.5 type embedding @@ -384,7 +381,7 @@ static bool encode_image_with_clip(clip_ctx * ctx_clip, int n_threads, const cli bool llava_validate_embed_size(const llama_context * ctx_llama, const clip_ctx * ctx_clip) { // make sure that the correct mmproj was used, i.e., compare apples to apples - int n_llama_embd = llama_n_embd(llama_get_model(ctx_llama)); + int n_llama_embd = llama_model_n_embd(llama_get_model(ctx_llama)); auto n_image_embd = clip_n_mmproj_embd(ctx_clip); if (n_image_embd != n_llama_embd) { LOG_ERR("%s: embedding dim of the multimodal projector (%d) is not equal to that of LLaMA (%d). Make sure that you use the correct mmproj file.\n", __func__, n_image_embd, n_llama_embd); @@ -456,7 +453,7 @@ struct llava_embd_batch { }; bool llava_eval_image_embed(llama_context * ctx_llama, const struct llava_image_embed * image_embed, int n_batch, int * n_past) { - int n_embd = llama_n_embd(llama_get_model(ctx_llama)); + int n_embd = llama_model_n_embd(llama_get_model(ctx_llama)); for (int i = 0; i < image_embed->n_image_pos; i += n_batch) { int n_eval = image_embed->n_image_pos - i; diff --git a/examples/llava/minicpmv-cli.cpp b/examples/llava/minicpmv-cli.cpp index e9cbb51ed..53d902d61 100644 --- a/examples/llava/minicpmv-cli.cpp +++ b/examples/llava/minicpmv-cli.cpp @@ -31,7 +31,7 @@ static struct llama_model * llava_init(common_params * params) { llama_model_params model_params = common_model_params_to_llama(*params); - llama_model * model = llama_load_model_from_file(params->model.c_str(), model_params); + llama_model * model = llama_model_load_from_file(params->model.c_str(), model_params); if (model == NULL) { LOG_ERR("%s: unable to load model\n" , __func__); return NULL; @@ -54,7 +54,7 @@ static struct llava_context * llava_init_context(common_params * params, llama_m ctx_params.n_ctx = params->n_ctx; } - llama_context * ctx_llama = llama_new_context_with_model(model, ctx_params); + llama_context * ctx_llama = llama_init_from_model(model, ctx_params); if (ctx_llama == NULL) { LOG_ERR("%s: failed to create the llama_context\n" , __func__); @@ -75,7 +75,7 @@ static void llava_free(struct llava_context * ctx_llava) { } llama_free(ctx_llava->ctx_llama); - llama_free_model(ctx_llava->model); + llama_model_free(ctx_llava->model); llama_backend_free(); } @@ -140,6 +140,9 @@ static void process_image(struct llava_context * ctx_llava, struct llava_image_e else if (has_minicpmv_projector == 3) { system_prompt = "<|im_start|>user\n"; } + else if (has_minicpmv_projector == 4) { + system_prompt = "<|im_start|>user\n"; + } LOG_INF("%s: image token past: %d\n", __func__, n_past); eval_string(ctx_llava->ctx_llama, (system_prompt+"").c_str(), params->n_batch, &n_past, false); process_eval_image_embed(ctx_llava, embeds, params->n_batch, &n_past, idx++); @@ -167,8 +170,12 @@ static const char * sample(struct common_sampler * smpl, int * n_past) { const llama_token id = common_sampler_sample(smpl, ctx_llama, -1); common_sampler_accept(smpl, id, true); + + const llama_model * model = llama_get_model(ctx_llama); + const llama_vocab * vocab = llama_model_get_vocab(model); + static std::string ret; - if (llama_token_is_eog(llama_get_model(ctx_llama), id)) { + if (llama_vocab_is_eog(vocab, id)) { ret = ""; } else { ret = common_token_to_piece(ctx_llama, id); @@ -223,6 +230,9 @@ static struct common_sampler * llama_init(struct llava_context * ctx_llava, comm else if (has_minicpmv_projector == 3) { user_prompt = "<|im_start|>user\n" + prompt; } + else if (has_minicpmv_projector == 4) { + user_prompt = "<|im_start|>user\n" + prompt; + } } eval_string(ctx_llava->ctx_llama, user_prompt.c_str(), params->n_batch, &n_past, false); @@ -232,6 +242,9 @@ static struct common_sampler * llama_init(struct llava_context * ctx_llava, comm else if (has_minicpmv_projector == 3) { eval_string(ctx_llava->ctx_llama, "<|im_end|><|im_start|>assistant\n", params->n_batch, &n_past, false); } + else if (has_minicpmv_projector == 4) { + eval_string(ctx_llava->ctx_llama, "<|im_end|><|im_start|>assistant\n", params->n_batch, &n_past, false); + } // generate the response @@ -304,7 +317,6 @@ int main(int argc, char ** argv) { const auto * tmp = llama_loop(ctx_llava, smpl, n_past); response += tmp; if (strcmp(tmp, "") == 0) break; - if (strstr(tmp, "###")) break; // Yi-VL behavior printf("%s", tmp);// mistral llava-1.6 if (strstr(response.c_str(), "")) break; // minicpm-v fflush(stdout); diff --git a/examples/llava/minicpmv-convert-image-encoder-to-gguf.py b/examples/llava/minicpmv-convert-image-encoder-to-gguf.py index ea773742a..9b196757f 100644 --- a/examples/llava/minicpmv-convert-image-encoder-to-gguf.py +++ b/examples/llava/minicpmv-convert-image-encoder-to-gguf.py @@ -501,7 +501,7 @@ default_image_mean = [0.48145466, 0.4578275, 0.40821073] default_image_std = [0.26862954, 0.26130258, 0.27577711] ap.add_argument('--image-mean', type=float, nargs='+', help='Mean of the images for normalization (overrides processor) ', default=None) ap.add_argument('--image-std', type=float, nargs='+', help='Standard deviation of the images for normalization (overrides processor)', default=None) -ap.add_argument('--minicpmv_version', type=int, help='minicpmv_version: MiniCPM-V-2 use 1; MiniCPM-V-2.5 use 2; MiniCPM-V-2.6 use 3', default=2) +ap.add_argument('--minicpmv_version', type=int, help='minicpmv_version: MiniCPM-V-2 use 1; MiniCPM-V-2.5 use 2; MiniCPM-V-2.6 use 3; MiniCPM-o-2.6 use 4', default=2) # with proper args = ap.parse_args() @@ -545,12 +545,19 @@ if args.use_f32: minicpmv_version = args.minicpmv_version emb_dim = 4096 +block_count = 26 if minicpmv_version == 1: emb_dim = 2304 + block_count = 26 elif minicpmv_version == 2: emb_dim = 4096 + block_count = 27 elif minicpmv_version == 3: emb_dim = 3584 + block_count = 27 +elif minicpmv_version == 4: + emb_dim = 3584 + block_count = 27 default_vision_config = { "hidden_size": 1152, @@ -567,6 +574,9 @@ model = Idefics2VisionTransformer(vision_config) if minicpmv_version == 3: vision_config = SiglipVisionConfig(**default_vision_config) model = SiglipVisionTransformer(vision_config) +elif minicpmv_version == 4: + vision_config = SiglipVisionConfig(**default_vision_config) + model = SiglipVisionTransformer(vision_config) processor = None # if model.attn_pool is not None: @@ -587,7 +597,7 @@ elif args.minicpmv_projector is not None: fname_middle = "mmproj-" has_text_encoder = False has_minicpmv_projector = True - minicpmv_version = 3 + minicpmv_version = 4 elif args.vision_only: fname_middle = "vision-" has_text_encoder = False @@ -625,7 +635,6 @@ if has_vision_encoder: fout.add_uint32("clip.vision.projection_dim", 0) fout.add_uint32(add_key_str(KEY_ATTENTION_HEAD_COUNT, VISION), 16) fout.add_float32(add_key_str(KEY_ATTENTION_LAYERNORM_EPS, VISION), 1e-6) - block_count = 26 fout.add_uint32(add_key_str(KEY_BLOCK_COUNT, VISION), block_count) if processor is not None: diff --git a/examples/llava/minicpmv-surgery.py b/examples/llava/minicpmv-surgery.py index 748ff5c57..ba8211658 100644 --- a/examples/llava/minicpmv-surgery.py +++ b/examples/llava/minicpmv-surgery.py @@ -8,7 +8,7 @@ ap.add_argument("-m", "--model", help="Path to MiniCPM-V model") args = ap.parse_args() # find the model part that includes the the multimodal projector weights -model = AutoModel.from_pretrained(args.model, trust_remote_code=True, local_files_only=True) +model = AutoModel.from_pretrained(args.model, trust_remote_code=True, local_files_only=True, torch_dtype=torch.bfloat16) checkpoint = model.state_dict() # get a list of mm tensor names diff --git a/examples/llava/qwen2vl-cli.cpp b/examples/llava/qwen2vl-cli.cpp index e86a60280..132a7da54 100644 --- a/examples/llava/qwen2vl-cli.cpp +++ b/examples/llava/qwen2vl-cli.cpp @@ -27,7 +27,7 @@ static bool qwen2vl_eval_image_embed(llama_context * ctx_llama, const struct llava_image_embed * image_embed, int n_batch, int * n_past, int * st_pos_id, struct clip_image_size * image_size) { - int n_embd = llama_n_embd(llama_get_model(ctx_llama)); + int n_embd = llama_model_n_embd(llama_get_model(ctx_llama)); const int patch_size = 14 * 2; const int ph = image_size->height / patch_size + (image_size->height % patch_size > 0); const int pw = image_size->width / patch_size + (image_size->width % patch_size > 0); @@ -132,8 +132,12 @@ static const char * sample(struct common_sampler * smpl, int * n_past, int * st_pos_id) { const llama_token id = common_sampler_sample(smpl, ctx_llama, -1); common_sampler_accept(smpl, id, true); + + const llama_model * model = llama_get_model(ctx_llama); + const llama_vocab * vocab = llama_model_get_vocab(model); + static std::string ret; - if (llama_token_is_eog(llama_get_model(ctx_llama), id)) { + if (llama_vocab_is_eog(vocab, id)) { ret = ""; } else { ret = common_token_to_piece(ctx_llama, id); @@ -310,7 +314,7 @@ static struct llama_model * llava_init(common_params * params) { llama_model_params model_params = common_model_params_to_llama(*params); - llama_model * model = llama_load_model_from_file(params->model.c_str(), model_params); + llama_model * model = llama_model_load_from_file(params->model.c_str(), model_params); if (model == NULL) { LOG_ERR("%s: unable to load model\n" , __func__); return NULL; @@ -328,11 +332,10 @@ static struct llava_context * llava_init_context(common_params * params, llama_m auto ctx_clip = clip_model_load(clip_path, /*verbosity=*/ 1); - llama_context_params ctx_params = common_context_params_to_llama(*params); ctx_params.n_ctx = params->n_ctx < 2048 ? 2048 : params->n_ctx; // we need a longer context size to process image embeddings - llama_context * ctx_llama = llama_new_context_with_model(model, ctx_params); + llama_context * ctx_llama = llama_init_from_model(model, ctx_params); if (ctx_llama == NULL) { LOG_ERR("%s: failed to create the llama_context\n" , __func__); @@ -354,7 +357,7 @@ static void llava_free(struct llava_context * ctx_llava) { } llama_free(ctx_llava->ctx_llama); - llama_free_model(ctx_llava->model); + llama_model_free(ctx_llava->model); llama_backend_free(); } @@ -481,7 +484,7 @@ static void debug_test_mrope_2d() { } static void debug_dump_img_embed(struct llava_context * ctx_llava) { - int n_embd = llama_n_embd(llama_get_model(ctx_llava->ctx_llama)); + int n_embd = llama_model_n_embd(llama_get_model(ctx_llava->ctx_llama)); int ne = n_embd * 4; float vals[56 * 56 * 3]; // float embd[ne]; @@ -575,7 +578,7 @@ int main(int argc, char ** argv) { } } - llama_free_model(model); + llama_model_free(model); return 0; } diff --git a/examples/lookahead/lookahead.cpp b/examples/lookahead/lookahead.cpp index e016618e3..2f0898e62 100644 --- a/examples/lookahead/lookahead.cpp +++ b/examples/lookahead/lookahead.cpp @@ -61,6 +61,8 @@ int main(int argc, char ** argv) { llama_model * model = llama_init.model.get(); llama_context * ctx = llama_init.context.get(); + const llama_vocab * vocab = llama_model_get_vocab(model); + // Tokenize the prompt std::vector inp; std::vector all; @@ -147,7 +149,7 @@ int main(int argc, char ** argv) { } // here we keep adding new n-grams as we go - ngram_container ngrams_observed(llama_n_vocab(model), N, G); + ngram_container ngrams_observed(llama_vocab_n_tokens(vocab), N, G); // debug struct llama_kv_cache_view kvc_view = llama_kv_cache_view_init(ctx, W + G + 1); @@ -297,7 +299,7 @@ int main(int argc, char ** argv) { } fflush(stdout); - if (llama_token_is_eog(model, id)) { + if (llama_vocab_is_eog(vocab, id)) { has_eos = true; } diff --git a/examples/lookup/lookup.cpp b/examples/lookup/lookup.cpp index 0d68b80b9..dbd0444ec 100644 --- a/examples/lookup/lookup.cpp +++ b/examples/lookup/lookup.cpp @@ -36,6 +36,8 @@ int main(int argc, char ** argv){ llama_model * model = llama_init.model.get(); llama_context * ctx = llama_init.context.get(); + const llama_vocab * vocab = llama_model_get_vocab(model); + // tokenize the prompt std::vector inp; inp = common_tokenize(ctx, params.prompt, true, true); @@ -136,7 +138,7 @@ int main(int argc, char ** argv){ LOG("%s", token_str.c_str()); } - if (llama_token_is_eog(model, id)) { + if (llama_vocab_is_eog(vocab, id)) { has_eos = true; } diff --git a/examples/main/README.md b/examples/main/README.md index 17d80a622..46f92eb7a 100644 --- a/examples/main/README.md +++ b/examples/main/README.md @@ -310,9 +310,9 @@ These options help improve the performance and memory usage of the LLaMA models. ### Batch Size -- `-b N, --batch-size N`: Set the batch size for prompt processing (default: `2048`). This large batch size benefits users who have BLAS installed and enabled it during the build. If you don't have BLAS enabled ("BLAS=0"), you can use a smaller number, such as 8, to see the prompt progress as it's evaluated in some situations. +- `-ub N`, `--ubatch-size N`: Physical batch size. This is the maximum number of tokens that may be processed at a time. Increasing this value may improve performance during prompt processing, at the expense of higher memory usage. Default: `512`. -- `-ub N`, `--ubatch-size N`: physical maximum batch size. This is for pipeline parallelization. Default: `512`. +- `-b N`, `--batch-size N`: Logical batch size. Increasing this value above the value of the physical batch size may improve prompt processing performance when using multiple GPUs with pipeline parallelism. Default: `2048`. ### Prompt Caching diff --git a/examples/main/main.cpp b/examples/main/main.cpp index b5e477f5b..da2a03ab9 100644 --- a/examples/main/main.cpp +++ b/examples/main/main.cpp @@ -4,8 +4,8 @@ #include "log.h" #include "sampling.h" #include "llama.h" +#include "chat-template.hpp" -#include #include #include #include @@ -31,6 +31,8 @@ #pragma warning(disable: 4244 4267) // possible loss of data #endif +static const char * DEFAULT_SYSTEM_MESSAGE = "You are a helpful assistant"; + static llama_context ** g_ctx; static llama_model ** g_model; static common_sampler ** g_smpl; @@ -83,14 +85,6 @@ static void sigint_handler(int signo) { } #endif -static std::string chat_add_and_format(struct llama_model * model, std::vector & chat_msgs, const std::string & role, const std::string & content) { - common_chat_msg new_msg{role, content}; - auto formatted = common_chat_format_single(model, g_params->chat_template, chat_msgs, new_msg, role == "user"); - chat_msgs.push_back({role, content}); - LOG_DBG("formatted: '%s'\n", formatted.c_str()); - return formatted; -} - int main(int argc, char ** argv) { common_params params; g_params = ¶ms; @@ -163,6 +157,9 @@ int main(int argc, char ** argv) { return 1; } + const llama_vocab * vocab = llama_model_get_vocab(model); + auto chat_templates = common_chat_templates_from_model(model, params.chat_template); + LOG_INF("%s: llama threadpool init, n_threads = %d\n", __func__, (int) params.cpuparams.n_threads); auto * reg = ggml_backend_dev_backend_reg(ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_CPU)); @@ -196,17 +193,33 @@ int main(int argc, char ** argv) { llama_attach_threadpool(ctx, threadpool, threadpool_batch); - const int n_ctx_train = llama_n_ctx_train(model); + const int n_ctx_train = llama_model_n_ctx_train(model); const int n_ctx = llama_n_ctx(ctx); if (n_ctx > n_ctx_train) { LOG_WRN("%s: model was trained on only %d context tokens (%d specified)\n", __func__, n_ctx_train, n_ctx); } + // auto enable conversation mode if chat template is available + const bool has_chat_template = chat_templates.has_explicit_template && chat_templates.template_default; + if (params.conversation_mode == COMMON_CONVERSATION_MODE_AUTO) { + if (has_chat_template) { + LOG_INF("%s: chat template is available, enabling conversation mode (disable it with -no-cnv)\n", __func__); + params.conversation_mode = COMMON_CONVERSATION_MODE_ENABLED; + } else { + params.conversation_mode = COMMON_CONVERSATION_MODE_DISABLED; + } + } + + // in case user force-activate conversation mode (via -cnv) without proper chat template, we show a warning + if (params.conversation_mode && !has_chat_template) { + LOG_WRN("%s: chat template is not available or is not supported. This may cause the model to output suboptimal responses\n", __func__); + } + // print chat template example in conversation mode - if (params.conversation) { + if (params.conversation_mode) { if (params.enable_chat_template) { - LOG_INF("%s: chat template example:\n%s\n", __func__, common_chat_format_example(model, params.chat_template).c_str()); + LOG_INF("%s: chat template example:\n%s\n", __func__, common_chat_format_example(*chat_templates.template_default, params.use_jinja).c_str()); } else { LOG_INF("%s: in-suffix/prefix is specified, chat template will be disabled\n", __func__); } @@ -241,18 +254,28 @@ int main(int argc, char ** argv) { } } - const bool add_bos = llama_add_bos_token(model); + const bool add_bos = llama_vocab_get_add_bos(vocab); if (!llama_model_has_encoder(model)) { - GGML_ASSERT(!llama_add_eos_token(model)); + GGML_ASSERT(!llama_vocab_get_add_eos(vocab)); } LOG_DBG("n_ctx: %d, add_bos: %d\n", n_ctx, add_bos); std::vector embd_inp; + auto chat_add_and_format = [&chat_msgs, &chat_templates](const std::string & role, const std::string & content) { + common_chat_msg new_msg{role, content}; + auto formatted = common_chat_format_single(*chat_templates.template_default, chat_msgs, new_msg, role == "user", g_params->use_jinja); + chat_msgs.push_back({role, content}); + LOG_DBG("formatted: '%s'\n", formatted.c_str()); + return formatted; + }; + { - auto prompt = (params.conversation && params.enable_chat_template && !params.prompt.empty()) - ? chat_add_and_format(model, chat_msgs, "system", params.prompt) // format the system prompt in conversation mode + auto prompt = (params.conversation_mode && params.enable_chat_template) + // format the system prompt in conversation mode (fallback to default if empty) + ? chat_add_and_format("system", params.prompt.empty() ? DEFAULT_SYSTEM_MESSAGE : params.prompt) + // otherwise use the prompt as is : params.prompt; if (params.interactive_first || !params.prompt.empty() || session_tokens.empty()) { LOG_DBG("tokenize the prompt\n"); @@ -269,7 +292,7 @@ int main(int argc, char ** argv) { // Should not run without any tokens if (embd_inp.empty()) { if (add_bos) { - embd_inp.push_back(llama_token_bos(model)); + embd_inp.push_back(llama_vocab_bos(vocab)); LOG_WRN("embd_inp was considered empty and bos was added: %s\n", string_from(ctx, embd_inp).c_str()); } else { LOG_ERR("input is empty\n"); @@ -326,7 +349,7 @@ int main(int argc, char ** argv) { params.n_keep += add_bos; // always keep the BOS token } - if (params.conversation) { + if (params.conversation_mode) { params.interactive_first = true; } @@ -450,7 +473,11 @@ int main(int argc, char ** argv) { #if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__)) || defined (_WIN32) LOG_INF( " - Press Ctrl+C to interject at any time.\n"); #endif - LOG_INF( "%s\n", control_message); + LOG_INF( "%s", control_message); + if (params.conversation_mode && params.enable_chat_template && params.prompt.empty()) { + LOG_INF( " - Using default system message. To change it, set a different value via -p PROMPT or -f FILE argument.\n"); + } + LOG_INF("\n"); is_interacting = params.interactive_first; } @@ -494,8 +521,8 @@ int main(int argc, char ** argv) { } llama_token decoder_start_token_id = llama_model_decoder_start_token(model); - if (decoder_start_token_id == -1) { - decoder_start_token_id = llama_token_bos(model); + if (decoder_start_token_id == LLAMA_TOKEN_NULL) { + decoder_start_token_id = llama_vocab_bos(vocab); } embd_inp.clear(); @@ -742,7 +769,7 @@ int main(int argc, char ** argv) { } // deal with end of generation tokens in interactive mode - if (llama_token_is_eog(model, common_sampler_last(smpl))) { + if (llama_vocab_is_eog(vocab, common_sampler_last(smpl))) { LOG_DBG("found an EOG token\n"); if (params.interactive) { @@ -754,7 +781,7 @@ int main(int argc, char ** argv) { } if (params.enable_chat_template) { - chat_add_and_format(model, chat_msgs, "assistant", assistant_ss.str()); + chat_add_and_format("assistant", assistant_ss.str()); } is_interacting = true; LOG("\n"); @@ -762,7 +789,7 @@ int main(int argc, char ** argv) { } // if current token is not EOG, we add it to current assistant message - if (params.conversation) { + if (params.conversation_mode) { const auto id = common_sampler_last(smpl); assistant_ss << common_token_to_piece(ctx, id, false); } @@ -770,17 +797,17 @@ int main(int argc, char ** argv) { if (n_past > 0 && is_interacting) { LOG_DBG("waiting for user input\n"); - if (params.conversation) { + if (params.conversation_mode) { LOG("\n> "); } if (params.input_prefix_bos) { LOG_DBG("adding input prefix BOS token\n"); - embd_inp.push_back(llama_token_bos(model)); + embd_inp.push_back(llama_vocab_bos(vocab)); } std::string buffer; - if (!params.input_prefix.empty() && !params.conversation) { + if (!params.input_prefix.empty() && !params.conversation_mode) { LOG_DBG("appending input prefix: '%s'\n", params.input_prefix.c_str()); LOG("%s", params.input_prefix.c_str()); } @@ -804,7 +831,7 @@ int main(int argc, char ** argv) { // Entering a empty line lets the user pass control back if (buffer.length() > 1) { // append input suffix if any - if (!params.input_suffix.empty() && !params.conversation) { + if (!params.input_suffix.empty() && !params.conversation_mode) { LOG_DBG("appending input suffix: '%s'\n", params.input_suffix.c_str()); LOG("%s", params.input_suffix.c_str()); } @@ -817,9 +844,9 @@ int main(int argc, char ** argv) { string_process_escapes(buffer); } - bool format_chat = params.conversation && params.enable_chat_template; + bool format_chat = params.conversation_mode && params.enable_chat_template; std::string user_inp = format_chat - ? chat_add_and_format(model, chat_msgs, "user", std::move(buffer)) + ? chat_add_and_format("user", std::move(buffer)) : std::move(buffer); // TODO: one inconvenient of current chat template implementation is that we can't distinguish between user input and special tokens (prefix/postfix) const auto line_pfx = common_tokenize(ctx, params.input_prefix, false, true); @@ -830,8 +857,8 @@ int main(int argc, char ** argv) { // if user stop generation mid-way, we must add EOT to finish model's last response if (need_insert_eot && format_chat) { - llama_token eot = llama_token_eot(model); - embd_inp.push_back(eot == -1 ? llama_token_eos(model) : eot); + llama_token eot = llama_vocab_eot(vocab); + embd_inp.push_back(eot == LLAMA_TOKEN_NULL ? llama_vocab_eos(vocab) : eot); need_insert_eot = false; } @@ -866,7 +893,7 @@ int main(int argc, char ** argv) { } // end of generation - if (!embd.empty() && llama_token_is_eog(model, embd.back()) && !(params.interactive)) { + if (!embd.empty() && llama_vocab_is_eog(vocab, embd.back()) && !(params.interactive)) { LOG(" [end of text]\n"); break; } diff --git a/examples/parallel/parallel.cpp b/examples/parallel/parallel.cpp index d48f51975..7ef43d5e1 100644 --- a/examples/parallel/parallel.cpp +++ b/examples/parallel/parallel.cpp @@ -135,6 +135,8 @@ int main(int argc, char ** argv) { llama_model * model = llama_init.model.get(); llama_context * ctx = llama_init.context.get(); + const llama_vocab * vocab = llama_model_get_vocab(model); + // load the prompts from an external file if there are any if (params.prompt.empty()) { LOG_INF("\033[32mNo new questions so proceed with build-in defaults.\033[0m\n"); @@ -358,7 +360,7 @@ int main(int argc, char ** argv) { // client.id, client.seq_id, id, client.n_decoded, client.i_batch, token_str.c_str()); if (client.n_decoded > 2 && - (llama_token_is_eog(model, id) || + (llama_vocab_is_eog(vocab, id) || (params.n_predict > 0 && client.n_decoded + client.n_prompt >= params.n_predict) || client.response.find("User:") != std::string::npos || client.response.find('\n') != std::string::npos)) { diff --git a/examples/passkey/passkey.cpp b/examples/passkey/passkey.cpp index 09bba708f..5953928d4 100644 --- a/examples/passkey/passkey.cpp +++ b/examples/passkey/passkey.cpp @@ -63,22 +63,24 @@ int main(int argc, char ** argv) { llama_model_params model_params = common_model_params_to_llama(params); - llama_model * model = llama_load_model_from_file(params.model.c_str(), model_params); + llama_model * model = llama_model_load_from_file(params.model.c_str(), model_params); if (model == NULL) { LOG_ERR("%s: unable to load model\n" , __func__); return 1; } + const llama_vocab * vocab = llama_model_get_vocab(model); + // initialize the context llama_context_params ctx_params = common_context_params_to_llama(params); - ctx_params.n_ctx = llama_n_ctx_train(model)*n_grp + n_keep; + ctx_params.n_ctx = llama_model_n_ctx_train(model)*n_grp + n_keep; GGML_ASSERT(ctx_params.n_batch % n_grp == 0 && "n_batch must be divisible by n_grp"); - llama_context * ctx = llama_new_context_with_model(model, ctx_params); + llama_context * ctx = llama_init_from_model(model, ctx_params); if (ctx == NULL) { LOG_ERR("%s: failed to create the llama_context\n" , __func__); return 1; @@ -223,7 +225,7 @@ int main(int argc, char ** argv) { const llama_token new_token_id = llama_sampler_sample(smpl, ctx, batch.n_tokens - 1); // is it an end of generation? - if (llama_token_is_eog(model, new_token_id) || n_cur == n_len) { + if (llama_vocab_is_eog(vocab, new_token_id) || n_cur == n_len) { LOG("\n"); break; @@ -266,7 +268,7 @@ int main(int argc, char ** argv) { llama_batch_free(batch); llama_free(ctx); - llama_free_model(model); + llama_model_free(model); llama_backend_free(); diff --git a/examples/perplexity/perplexity.cpp b/examples/perplexity/perplexity.cpp index 6bdc57f8e..9bf6c5743 100644 --- a/examples/perplexity/perplexity.cpp +++ b/examples/perplexity/perplexity.cpp @@ -296,8 +296,11 @@ static results_perplexity perplexity_v2(llama_context * ctx, const common_params // Output: `perplexity: 13.5106 [114/114]` // BOS tokens will be added for each chunk before eval - const bool add_bos = llama_add_bos_token(llama_get_model(ctx)); - GGML_ASSERT(!llama_add_eos_token(llama_get_model(ctx))); + const llama_model * model = llama_get_model(ctx); + const llama_vocab * vocab = llama_model_get_vocab(model); + + const bool add_bos = llama_vocab_get_add_bos(vocab); + GGML_ASSERT(!llama_vocab_get_add_eos(vocab)); LOG_INF("%s: tokenizing the input ..\n", __func__); @@ -338,7 +341,7 @@ static results_perplexity perplexity_v2(llama_context * ctx, const common_params const int n_chunk = params.n_chunks < 0 ? n_chunk_max : std::min(params.n_chunks, n_chunk_max); const int n_batch = params.n_batch; - const int n_vocab = llama_n_vocab(llama_get_model(ctx)); + const int n_vocab = llama_vocab_n_tokens(vocab); int count = 0; double nll = 0.0; @@ -382,7 +385,7 @@ static results_perplexity perplexity_v2(llama_context * ctx, const common_params // add BOS token for the first batch of each chunk if (add_bos && j == 0) { - tokens[batch_start] = llama_token_bos(llama_get_model(ctx)); + tokens[batch_start] = llama_vocab_bos(vocab); } const auto * batch_logits = llama_get_logits(ctx); @@ -444,8 +447,11 @@ static results_perplexity perplexity(llama_context * ctx, const common_params & // Output: `perplexity: 13.5106 [114/114]` // BOS tokens will be added for each chunk before eval - const bool add_bos = llama_add_bos_token(llama_get_model(ctx)); - GGML_ASSERT(!llama_add_eos_token(llama_get_model(ctx))); + const llama_model * model = llama_get_model(ctx); + const llama_vocab * vocab = llama_model_get_vocab(model); + + const bool add_bos = llama_vocab_get_add_bos(vocab); + GGML_ASSERT(!llama_vocab_get_add_eos(vocab)); std::ofstream logits_stream; if (!params.logits_file.empty()) { @@ -485,7 +491,7 @@ static results_perplexity perplexity(llama_context * ctx, const common_params & const int n_chunk = params.n_chunks < 0 ? n_chunk_max : std::min(params.n_chunks, n_chunk_max); const int n_batch = params.n_batch; - const int n_vocab = llama_n_vocab(llama_get_model(ctx)); + const int n_vocab = llama_vocab_n_tokens(vocab); int count = 0; double nll = 0.0; @@ -557,7 +563,7 @@ static results_perplexity perplexity(llama_context * ctx, const common_params & // add BOS token for the first batch of each chunk if (add_bos && j == 0) { - tokens[seq_start] = llama_token_bos(llama_get_model(ctx)); + tokens[seq_start] = llama_vocab_bos(vocab); } for (int k = 0; k < batch_size; ++k) { @@ -732,6 +738,9 @@ static void compute_logprobs(const float * batch_logits, int n_vocab, std::vecto } static void hellaswag_score(llama_context * ctx, const common_params & params) { + const llama_model * model = llama_get_model(ctx); + const llama_vocab * vocab = llama_model_get_vocab(model); + // Calculates hellaswag score (acc_norm) from prompt // // Data extracted from the HellaSwag validation dataset (MIT license) https://github.com/rowanz/hellaswag/blob/master/data/hellaswag_val.jsonl @@ -765,7 +774,7 @@ static void hellaswag_score(llama_context * ctx, const common_params & params) { size_t hs_task_count = prompt_lines.size()/6; LOG_INF("%s : loaded %zu tasks from prompt.\n", __func__, hs_task_count); - const bool is_spm = llama_vocab_type(llama_get_model(ctx)) == LLAMA_VOCAB_TYPE_SPM; + const bool is_spm = llama_vocab_type(vocab) == LLAMA_VOCAB_TYPE_SPM; LOG_INF("================================= is_spm = %d\n", is_spm); // The tasks should be randomized so the score stabilizes quickly. @@ -848,7 +857,7 @@ static void hellaswag_score(llama_context * ctx, const common_params & params) { const int n_ctx = llama_n_ctx(ctx); const int n_batch = params.n_batch; - const int n_vocab = llama_n_vocab(llama_get_model(ctx)); + const int n_vocab = llama_vocab_n_tokens(vocab); const int max_tasks_per_batch = 32; const int max_seq = std::min(4*max_tasks_per_batch, (int) llama_n_seq_max(ctx)); @@ -1072,6 +1081,8 @@ static std::vector load_winogrande_from_csv(const std::string * */ static void winogrande_score(llama_context * ctx, const common_params & params) { + const llama_model * model = llama_get_model(ctx); + const llama_vocab * vocab = llama_model_get_vocab(model); constexpr int k_min_trailing_ctx = 3; @@ -1130,7 +1141,7 @@ static void winogrande_score(llama_context * ctx, const common_params & params) const int n_ctx = llama_n_ctx(ctx); const int n_batch = params.n_batch; - const int n_vocab = llama_n_vocab(llama_get_model(ctx)); + const int n_vocab = llama_vocab_n_tokens(vocab); const int max_tasks_per_batch = 128; const int max_seq = std::min(2*max_tasks_per_batch, (int) llama_n_seq_max(ctx)); @@ -1374,6 +1385,8 @@ static bool multiple_choice_prepare_one_task(llama_context * ctx, multiple_choic // https://huggingface.co/datasets/truthful_qa // static void multiple_choice_score(llama_context * ctx, const common_params & params) { + const llama_model * model = llama_get_model(ctx); + const llama_vocab * vocab = llama_model_get_vocab(model); std::istringstream strstream(params.prompt); uint32_t n_task; @@ -1482,7 +1495,7 @@ static void multiple_choice_score(llama_context * ctx, const common_params & par const int n_ctx = llama_n_ctx(ctx); const int n_batch = params.n_batch; - const int n_vocab = llama_n_vocab(llama_get_model(ctx)); + const int n_vocab = llama_vocab_n_tokens(vocab); const int max_tasks_per_batch = 32; const int max_seq = std::min(4*max_tasks_per_batch, (int) llama_n_seq_max(ctx)); @@ -1655,6 +1668,9 @@ static void multiple_choice_score(llama_context * ctx, const common_params & par } static void kl_divergence(llama_context * ctx, const common_params & params) { + const llama_model * model = llama_get_model(ctx); + const llama_vocab * vocab = llama_model_get_vocab(model); + if (params.logits_file.empty()) { LOG_ERR("%s: you must provide a name of a file containing the log probabilities of the base model\n", __func__); return; @@ -1688,8 +1704,8 @@ static void kl_divergence(llama_context * ctx, const common_params & params) { LOG_ERR("%s: failed reading n_vocab, n_chunk from %s\n", __func__, params.logits_file.c_str()); return; } - if (n_vocab != llama_n_vocab(llama_get_model(ctx))) { - LOG_ERR("%s: inconsistent vocabulary (%d vs %d)\n", __func__, n_vocab, llama_n_vocab(llama_get_model(ctx))); + if (n_vocab != llama_vocab_n_tokens(vocab)) { + LOG_ERR("%s: inconsistent vocabulary (%d vs %d)\n", __func__, n_vocab, llama_vocab_n_tokens(vocab)); } std::vector tokens(size_t(n_ctx) * n_chunk); @@ -1701,8 +1717,8 @@ static void kl_divergence(llama_context * ctx, const common_params & params) { const int n_batch = params.n_batch; const int num_batches = (n_ctx + n_batch - 1)/n_batch; const int nv = 2*((n_vocab + 1)/2) + 4; - const bool add_bos = llama_add_bos_token(llama_get_model(ctx)); - GGML_ASSERT(!llama_add_eos_token(llama_get_model(ctx))); + const bool add_bos = llama_vocab_get_add_bos(vocab); + GGML_ASSERT(!llama_vocab_get_add_eos(vocab)); std::vector log_probs_uint16(size_t(n_ctx - 1 - n_ctx/2) * nv); std::vector kld_values(size_t(n_ctx - 1 - n_ctx/2)*n_chunk); @@ -1761,7 +1777,7 @@ static void kl_divergence(llama_context * ctx, const common_params & params) { // add BOS token for the first batch of each chunk if (add_bos && j == 0) { - tokens[batch_start] = llama_token_bos(llama_get_model(ctx)); + tokens[batch_start] = llama_vocab_bos(vocab); } common_batch_clear(batch); @@ -1995,7 +2011,7 @@ int main(int argc, char ** argv) { return 1; } - const int n_ctx_train = llama_n_ctx_train(model); + const int n_ctx_train = llama_model_n_ctx_train(model); if (params.n_ctx > n_ctx_train) { LOG_WRN("%s: model was trained on only %d context tokens (%d specified)\n", diff --git a/examples/quantize-stats/quantize-stats.cpp b/examples/quantize-stats/quantize-stats.cpp index ab91d0b40..bd2f73467 100644 --- a/examples/quantize-stats/quantize-stats.cpp +++ b/examples/quantize-stats/quantize-stats.cpp @@ -309,7 +309,7 @@ int main(int argc, char ** argv) { auto mparams = llama_model_default_params(); mparams.use_mlock = false; - model = llama_load_model_from_file(params.model.c_str(), mparams); + model = llama_model_load_from_file(params.model.c_str(), mparams); if (model == NULL) { fprintf(stderr, "%s: error: failed to load model '%s'\n", __func__, params.model.c_str()); @@ -319,11 +319,11 @@ int main(int argc, char ** argv) { auto cparams = llama_context_default_params(); cparams.n_ctx = 256; - ctx = llama_new_context_with_model(model, cparams); + ctx = llama_init_from_model(model, cparams); if (ctx == NULL) { fprintf(stderr, "%s: error: failed to create context with model '%s'\n", __func__, params.model.c_str()); - llama_free_model(model); + llama_model_free(model); return 1; } } @@ -347,7 +347,7 @@ int main(int argc, char ** argv) { fprintf(stderr, "%s: error: Quantization should be tested with a float model, " "this model contains already quantized layers (%s is type %d)\n", __func__, kv_tensor.first.c_str(), kv_tensor.second->type); llama_free(ctx); - llama_free_model(model); + llama_model_free(model); return 1; } included_layers++; @@ -409,7 +409,7 @@ int main(int argc, char ** argv) { llama_free(ctx); - llama_free_model(model); + llama_model_free(model); // report timing { const int64_t t_main_end_us = ggml_time_us(); diff --git a/examples/quantize/tests.sh b/examples/quantize/tests.sh index 24bc970e8..70f7610f9 100644 --- a/examples/quantize/tests.sh +++ b/examples/quantize/tests.sh @@ -47,7 +47,7 @@ echo PASS echo # 3a. Test the requanted model is loading properly -$MAIN --model $WORK_PATH/ggml-model-requant-00001-of-00006.gguf --n-predict 32 +$MAIN -no-cnv --model $WORK_PATH/ggml-model-requant-00001-of-00006.gguf --n-predict 32 echo PASS echo @@ -57,7 +57,7 @@ echo PASS echo # 4b. Test the requanted model is loading properly -$MAIN --model $WORK_PATH/ggml-model-requant-merge.gguf --n-predict 32 +$MAIN -no-cnv --model $WORK_PATH/ggml-model-requant-merge.gguf --n-predict 32 echo PASS echo diff --git a/examples/retrieval/retrieval.cpp b/examples/retrieval/retrieval.cpp index f534b5eff..2439022a2 100644 --- a/examples/retrieval/retrieval.cpp +++ b/examples/retrieval/retrieval.cpp @@ -159,7 +159,9 @@ int main(int argc, char ** argv) { return 1; } - const int n_ctx_train = llama_n_ctx_train(model); + const llama_vocab * vocab = llama_model_get_vocab(model); + + const int n_ctx_train = llama_model_n_ctx_train(model); const int n_ctx = llama_n_ctx(ctx); const enum llama_pooling_type pooling_type = llama_pooling_type(ctx); @@ -192,8 +194,8 @@ int main(int argc, char ** argv) { return 1; } // add eos if not present - if (llama_token_eos(model) >= 0 && (inp.empty() || inp.back() != llama_token_eos(model))) { - inp.push_back(llama_token_eos(model)); + if (llama_vocab_eos(vocab) >= 0 && (inp.empty() || inp.back() != llama_vocab_eos(vocab))) { + inp.push_back(llama_vocab_eos(vocab)); } chunk.tokens = inp; } @@ -215,7 +217,7 @@ int main(int argc, char ** argv) { struct llama_batch batch = llama_batch_init(n_batch, 0, 1); // allocate output - const int n_embd = llama_n_embd(model); + const int n_embd = llama_model_n_embd(model); std::vector embeddings(n_chunks * n_embd, 0); float * emb = embeddings.data(); diff --git a/examples/run/CMakeLists.txt b/examples/run/CMakeLists.txt index 0686d6305..cd6b0520e 100644 --- a/examples/run/CMakeLists.txt +++ b/examples/run/CMakeLists.txt @@ -1,5 +1,5 @@ set(TARGET llama-run) -add_executable(${TARGET} run.cpp) +add_executable(${TARGET} run.cpp linenoise.cpp/linenoise.cpp) install(TARGETS ${TARGET} RUNTIME) target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT}) target_compile_features(${TARGET} PRIVATE cxx_std_17) diff --git a/examples/run/README.md b/examples/run/README.md index a06805441..89a552079 100644 --- a/examples/run/README.md +++ b/examples/run/README.md @@ -3,11 +3,10 @@ The purpose of this example is to demonstrate a minimal usage of llama.cpp for running models. ```bash -llama-run granite-code +llama-run granite3-moe ``` ```bash -llama-run -h Description: Runs a llm @@ -17,7 +16,7 @@ Usage: Options: -c, --context-size Context size (default: 2048) - -n, --ngl + -n, -ngl, --ngl Number of GPU layers (default: 0) --temp Temperature (default: 0.8) diff --git a/examples/run/linenoise.cpp/LICENSE b/examples/run/linenoise.cpp/LICENSE new file mode 100644 index 000000000..b006b3b24 --- /dev/null +++ b/examples/run/linenoise.cpp/LICENSE @@ -0,0 +1,26 @@ +Copyright (c) 2010-2014, Salvatore Sanfilippo +Copyright (c) 2010-2013, Pieter Noordhuis +Copyright (c) 2025, Eric Curtin + +All rights reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are met: + +* Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + +* Redistributions in binary form must reproduce the above copyright notice, + this list of conditions and the following disclaimer in the documentation + and/or other materials provided with the distribution. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND +ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED +WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR +ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES +(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; +LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON +ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS +SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. diff --git a/examples/run/linenoise.cpp/linenoise.cpp b/examples/run/linenoise.cpp/linenoise.cpp new file mode 100644 index 000000000..a68f12a1a --- /dev/null +++ b/examples/run/linenoise.cpp/linenoise.cpp @@ -0,0 +1,1350 @@ +#ifndef _WIN32 +/* + * You can find the latest source code at: + * + * http://github.com/ericcurtin/linenoise.cpp + * + * Does a number of crazy assumptions that happen to be true in 99.9999% of + * the 2010 UNIX computers around. + * + * ------------------------------------------------------------------------ + * + * Copyright (c) 2010-2023, Salvatore Sanfilippo + * Copyright (c) 2010-2013, Pieter Noordhuis + * Copyright (c) 2025, Eric Curtin + * + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are + * met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * + * ------------------------------------------------------------------------ + * + * References: + * - http://invisible-island.net/xterm/ctlseqs/ctlseqs.html + * - http://www.3waylabs.com/nw/WWW/products/wizcon/vt220.html + * + * Todo list: + * - Filter bogus Ctrl+ combinations. + * - Win32 support + * + * Bloat: + * - History search like Ctrl+r in readline? + * + * List of escape sequences used by this program, we do everything just + * with three sequences. In order to be so cheap we may have some + * flickering effect with some slow terminal, but the lesser sequences + * the more compatible. + * + * EL (Erase Line) + * Sequence: ESC [ n K + * Effect: if n is 0 or missing, clear from cursor to end of line + * Effect: if n is 1, clear from beginning of line to cursor + * Effect: if n is 2, clear entire line + * + * CUF (CUrsor Forward) + * Sequence: ESC [ n C + * Effect: moves cursor forward n chars + * + * CUB (CUrsor Backward) + * Sequence: ESC [ n D + * Effect: moves cursor backward n chars + * + * The following is used to get the terminal width if getting + * the width with the TIOCGWINSZ ioctl fails + * + * DSR (Device Status Report) + * Sequence: ESC [ 6 n + * Effect: reports the current cusor position as ESC [ n ; m R + * where n is the row and m is the column + * + * When multi line mode is enabled, we also use an additional escape + * sequence. However multi line editing is disabled by default. + * + * CUU (Cursor Up) + * Sequence: ESC [ n A + * Effect: moves cursor up of n chars. + * + * CUD (Cursor Down) + * Sequence: ESC [ n B + * Effect: moves cursor down of n chars. + * + * When linenoiseClearScreen() is called, two additional escape sequences + * are used in order to clear the screen and position the cursor at home + * position. + * + * CUP (Cursor position) + * Sequence: ESC [ H + * Effect: moves the cursor to upper left corner + * + * ED (Erase display) + * Sequence: ESC [ 2 J + * Effect: clear the whole screen + * + */ + +# include "linenoise.h" + +# include +# include +# include +# include +# include +# include +# include +# include +# include +# include + +# include +# include +# include + +# define LINENOISE_DEFAULT_HISTORY_MAX_LEN 100 +# define LINENOISE_MAX_LINE 4096 +static std::vector unsupported_term = { "dumb", "cons25", "emacs" }; +static linenoiseCompletionCallback *completionCallback = NULL; +static linenoiseHintsCallback *hintsCallback = NULL; +static linenoiseFreeHintsCallback *freeHintsCallback = NULL; +static char *linenoiseNoTTY(void); +static void refreshLineWithCompletion(struct linenoiseState *ls, linenoiseCompletions *lc, int flags); +static void refreshLineWithFlags(struct linenoiseState *l, int flags); + +static struct termios orig_termios; /* In order to restore at exit.*/ +static int maskmode = 0; /* Show "***" instead of input. For passwords. */ +static int rawmode = 0; /* For atexit() function to check if restore is needed*/ +static int mlmode = 0; /* Multi line mode. Default is single line. */ +static int atexit_registered = 0; /* Register atexit just 1 time. */ +static int history_max_len = LINENOISE_DEFAULT_HISTORY_MAX_LEN; +static int history_len = 0; +static char **history = NULL; + +enum KEY_ACTION{ + KEY_NULL = 0, /* NULL */ + CTRL_A = 1, /* Ctrl+a */ + CTRL_B = 2, /* Ctrl-b */ + CTRL_C = 3, /* Ctrl-c */ + CTRL_D = 4, /* Ctrl-d */ + CTRL_E = 5, /* Ctrl-e */ + CTRL_F = 6, /* Ctrl-f */ + CTRL_H = 8, /* Ctrl-h */ + TAB = 9, /* Tab */ + CTRL_K = 11, /* Ctrl+k */ + CTRL_L = 12, /* Ctrl+l */ + ENTER = 13, /* Enter */ + CTRL_N = 14, /* Ctrl-n */ + CTRL_P = 16, /* Ctrl-p */ + CTRL_T = 20, /* Ctrl-t */ + CTRL_U = 21, /* Ctrl+u */ + CTRL_W = 23, /* Ctrl+w */ + ESC = 27, /* Escape */ + BACKSPACE = 127 /* Backspace */ +}; + +static void linenoiseAtExit(void); +int linenoiseHistoryAdd(const char *line); +#define REFRESH_CLEAN (1<<0) // Clean the old prompt from the screen +#define REFRESH_WRITE (1<<1) // Rewrite the prompt on the screen. +#define REFRESH_ALL (REFRESH_CLEAN|REFRESH_WRITE) // Do both. +static void refreshLine(struct linenoiseState *l); + +class File { + public: + FILE * file = nullptr; + + FILE * open(const std::string & filename, const char * mode) { + file = fopen(filename.c_str(), mode); + + return file; + } + + int lock() { + if (file) { + fd = fileno(file); + if (flock(fd, LOCK_EX | LOCK_NB) != 0) { + fd = -1; + + return 1; + } + } + + return 0; + } + + ~File() { + if (fd >= 0) { + flock(fd, LOCK_UN); + } + + if (file) { + fclose(file); + } + } + + private: + int fd = -1; +}; + +__attribute__((format(printf, 1, 2))) +/* Debugging function. */ +#if 0 +static void lndebug(const char *fmt, ...) { + static File file; + if (file.file == nullptr) { + file.open("/tmp/lndebug.txt", "a"); + } + + if (file.file != nullptr) { + va_list args; + va_start(args, fmt); + vfprintf(file.file, fmt, args); + va_end(args); + fflush(file.file); + } +} +#else +static void lndebug(const char *, ...) { +} +#endif + +/* ======================= Low level terminal handling ====================== */ + +/* Enable "mask mode". When it is enabled, instead of the input that + * the user is typing, the terminal will just display a corresponding + * number of asterisks, like "****". This is useful for passwords and other + * secrets that should not be displayed. */ +void linenoiseMaskModeEnable(void) { + maskmode = 1; +} + +/* Disable mask mode. */ +void linenoiseMaskModeDisable(void) { + maskmode = 0; +} + +/* Set if to use or not the multi line mode. */ +void linenoiseSetMultiLine(int ml) { + mlmode = ml; +} + +/* Return true if the terminal name is in the list of terminals we know are + * not able to understand basic escape sequences. */ +static int isUnsupportedTerm(void) { + char *term = getenv("TERM"); + if (term == NULL) return 0; + for (size_t j = 0; j < unsupported_term.size(); ++j) { + if (!strcasecmp(term, unsupported_term[j])) { + return 1; + } + } + return 0; +} + +/* Raw mode: 1960 magic shit. */ +static int enableRawMode(int fd) { + struct termios raw; + + if (!isatty(STDIN_FILENO)) goto fatal; + if (!atexit_registered) { + atexit(linenoiseAtExit); + atexit_registered = 1; + } + if (tcgetattr(fd,&orig_termios) == -1) goto fatal; + + raw = orig_termios; /* modify the original mode */ + /* input modes: no break, no CR to NL, no parity check, no strip char, + * no start/stop output control. */ + raw.c_iflag &= ~(BRKINT | ICRNL | INPCK | ISTRIP | IXON); + /* output modes - disable post processing */ + raw.c_oflag &= ~(OPOST); + /* control modes - set 8 bit chars */ + raw.c_cflag |= (CS8); + /* local modes - choing off, canonical off, no extended functions, + * no signal chars (^Z,^C) */ + raw.c_lflag &= ~(ECHO | ICANON | IEXTEN | ISIG); + /* control chars - set return condition: min number of bytes and timer. + * We want read to return every single byte, without timeout. */ + raw.c_cc[VMIN] = 1; raw.c_cc[VTIME] = 0; /* 1 byte, no timer */ + + /* put terminal in raw mode after flushing */ + if (tcsetattr(fd,TCSAFLUSH,&raw) < 0) goto fatal; + rawmode = 1; + return 0; + +fatal: + errno = ENOTTY; + return -1; +} + +static void disableRawMode(int fd) { + /* Don't even check the return value as it's too late. */ + if (rawmode && tcsetattr(fd,TCSAFLUSH,&orig_termios) != -1) + rawmode = 0; +} + +/* Use the ESC [6n escape sequence to query the horizontal cursor position + * and return it. On error -1 is returned, on success the position of the + * cursor. */ +static int getCursorPosition(int ifd, int ofd) { + char buf[32]; + int cols, rows; + unsigned int i = 0; + + /* Report cursor location */ + if (write(ofd, "\x1b[6n", 4) != 4) return -1; + + /* Read the response: ESC [ rows ; cols R */ + while (i < sizeof(buf)-1) { + if (read(ifd,buf+i,1) != 1) break; + if (buf[i] == 'R') break; + i++; + } + buf[i] = '\0'; + + /* Parse it. */ + if (buf[0] != ESC || buf[1] != '[') return -1; + if (sscanf(buf+2,"%d;%d",&rows,&cols) != 2) return -1; + return cols; +} + +/* Try to get the number of columns in the current terminal, or assume 80 + * if it fails. */ +static int getColumns(int ifd, int ofd) { + struct winsize ws; + + if (ioctl(1, TIOCGWINSZ, &ws) == -1 || ws.ws_col == 0) { + /* ioctl() failed. Try to query the terminal itself. */ + int start, cols; + + /* Get the initial position so we can restore it later. */ + start = getCursorPosition(ifd,ofd); + if (start == -1) goto failed; + + /* Go to right margin and get position. */ + if (write(ofd,"\x1b[999C",6) != 6) goto failed; + cols = getCursorPosition(ifd,ofd); + if (cols == -1) goto failed; + + /* Restore position. */ + if (cols > start) { + char seq[32]; + snprintf(seq,32,"\x1b[%dD",cols-start); + if (write(ofd,seq,strlen(seq)) == -1) { + /* Can't recover... */ + } + } + return cols; + } else { + return ws.ws_col; + } + +failed: + return 80; +} + +/* Clear the screen. Used to handle ctrl+l */ +void linenoiseClearScreen(void) { + if (write(STDOUT_FILENO,"\x1b[H\x1b[2J",7) <= 0) { + /* nothing to do, just to avoid warning. */ + } +} + +/* Beep, used for completion when there is nothing to complete or when all + * the choices were already shown. */ +static void linenoiseBeep(void) { + fprintf(stderr, "\x7"); + fflush(stderr); +} + +/* Called by completeLine() and linenoiseShow() to render the current + * edited line with the proposed completion. If the current completion table + * is already available, it is passed as second argument, otherwise the + * function will use the callback to obtain it. + * + * Flags are the same as refreshLine*(), that is REFRESH_* macros. */ +static void refreshLineWithCompletion(struct linenoiseState *ls, linenoiseCompletions *lc, int flags) { + /* Obtain the table of completions if the caller didn't provide one. */ + linenoiseCompletions ctable; + if (lc == NULL) { + completionCallback(ls->buf, &ctable); + lc = &ctable; + } + + /* Show the edited line with completion if possible, or just refresh. */ + if (ls->completion_idx < lc->len) { + struct linenoiseState saved = *ls; + ls->len = ls->pos = strlen(lc->cvec[ls->completion_idx]); + ls->buf = lc->cvec[ls->completion_idx]; + refreshLineWithFlags(ls, flags); + ls->len = saved.len; + ls->pos = saved.pos; + ls->buf = saved.buf; + } else { + refreshLineWithFlags(ls, flags); + } + + if (lc == &ctable) { + ctable.to_free = false; + } +} + +/* This is an helper function for linenoiseEdit*() and is called when the + * user types the key in order to complete the string currently in the + * input. + * + * The state of the editing is encapsulated into the pointed linenoiseState + * structure as described in the structure definition. + * + * If the function returns non-zero, the caller should handle the + * returned value as a byte read from the standard input, and process + * it as usually: this basically means that the function may return a byte + * read from the termianl but not processed. Otherwise, if zero is returned, + * the input was consumed by the completeLine() function to navigate the + * possible completions, and the caller should read for the next characters + * from stdin. */ +static int completeLine(struct linenoiseState *ls, int keypressed) { + linenoiseCompletions lc; + int nwritten; + char c = keypressed; + + completionCallback(ls->buf, &lc); + if (lc.len == 0) { + linenoiseBeep(); + ls->in_completion = 0; + } else { + switch(c) { + case 9: /* tab */ + if (ls->in_completion == 0) { + ls->in_completion = 1; + ls->completion_idx = 0; + } else { + ls->completion_idx = (ls->completion_idx + 1) % (lc.len + 1); + if (ls->completion_idx == lc.len) linenoiseBeep(); + } + c = 0; + break; + case 27: /* escape */ + /* Re-show original buffer */ + if (ls->completion_idx < lc.len) refreshLine(ls); + ls->in_completion = 0; + c = 0; + break; + default: + /* Update buffer and return */ + if (ls->completion_idx < lc.len) { + nwritten = snprintf(ls->buf, ls->buflen, "%s", lc.cvec[ls->completion_idx]); + ls->len = ls->pos = nwritten; + } + ls->in_completion = 0; + break; + } + + /* Show completion or original buffer */ + if (ls->in_completion && ls->completion_idx < lc.len) { + refreshLineWithCompletion(ls, &lc, REFRESH_ALL); + } else { + refreshLine(ls); + } + } + + return c; /* Return last read character */ +} + +/* Register a callback function to be called for tab-completion. */ +void linenoiseSetCompletionCallback(linenoiseCompletionCallback *fn) { + completionCallback = fn; +} + +/* Register a hits function to be called to show hits to the user at the + * right of the prompt. */ +void linenoiseSetHintsCallback(linenoiseHintsCallback *fn) { + hintsCallback = fn; +} + +/* Register a function to free the hints returned by the hints callback + * registered with linenoiseSetHintsCallback(). */ +void linenoiseSetFreeHintsCallback(linenoiseFreeHintsCallback *fn) { + freeHintsCallback = fn; +} + +/* This function is used by the callback function registered by the user + * in order to add completion options given the input string when the + * user typed . See the example.c source code for a very easy to + * understand example. */ +void linenoiseAddCompletion(linenoiseCompletions *lc, const char *str) { + const size_t len = strlen(str); + auto copy = std::make_unique(len + 1); + if (!copy) { + return; + } + + memcpy(copy.get(), str, len + 1); + char ** cvec = static_cast(std::realloc(lc->cvec, sizeof(char *) * (lc->len + 1))); + if (cvec == nullptr) { + return; + } + + lc->cvec = cvec; + lc->cvec[lc->len++] = copy.release(); +} + +/* Helper of refreshSingleLine() and refreshMultiLine() to show hints + * to the right of the prompt. */ +static void refreshShowHints(std::string & ab, struct linenoiseState * l, int plen) { + char seq[64]; + if (hintsCallback && plen+l->len < l->cols) { + int color = -1, bold = 0; + const char *hint = hintsCallback(l->buf,&color,&bold); + if (hint) { + int hintlen = strlen(hint); + int hintmaxlen = l->cols-(plen+l->len); + if (hintlen > hintmaxlen) hintlen = hintmaxlen; + if (bold == 1 && color == -1) color = 37; + if (color != -1 || bold != 0) + snprintf(seq,64,"\033[%d;%d;49m",bold,color); + else + seq[0] = '\0'; + ab.append(seq); + ab.append(hint, hintlen); + if (color != -1 || bold != 0) + ab.append("\033[0m"); + + /* Call the function to free the hint returned. */ + if (freeHintsCallback) freeHintsCallback(hint); + } + } +} + +/* Single line low level line refresh. + * + * Rewrite the currently edited line accordingly to the buffer content, + * cursor position, and number of columns of the terminal. + * + * Flags is REFRESH_* macros. The function can just remove the old + * prompt, just write it, or both. */ +static void refreshSingleLine(struct linenoiseState *l, int flags) { + char seq[64]; + size_t plen = strlen(l->prompt); + int fd = l->ofd; + char *buf = l->buf; + size_t len = l->len; + size_t pos = l->pos; + std::string ab; + while((plen+pos) >= l->cols) { + buf++; + len--; + pos--; + } + while (plen+len > l->cols) { + len--; + } + + /* Cursor to left edge */ + snprintf(seq,sizeof(seq),"\r"); + ab.append(seq); + + if (flags & REFRESH_WRITE) { + /* Write the prompt and the current buffer content */ + ab.append(l->prompt); + if (maskmode == 1) { + while (len--) { + ab.append("*"); + } + } else { + ab.append(buf, len); + } + /* Show hits if any. */ + refreshShowHints(ab, l, plen); + } + + /* Erase to right */ + snprintf(seq,sizeof(seq),"\x1b[0K"); + ab.append(seq); + if (flags & REFRESH_WRITE) { + /* Move cursor to original position. */ + snprintf(seq,sizeof(seq),"\r\x1b[%dC", (int)(pos+plen)); + ab.append(seq); + } + + (void) !write(fd, ab.c_str(), ab.size()); /* Can't recover from write error. */ +} + +/* Multi line low level line refresh. + * + * Rewrite the currently edited line accordingly to the buffer content, + * cursor position, and number of columns of the terminal. + * + * Flags is REFRESH_* macros. The function can just remove the old + * prompt, just write it, or both. */ +static void refreshMultiLine(struct linenoiseState *l, int flags) { + char seq[64]; + int plen = strlen(l->prompt); + int rows = (plen+l->len+l->cols-1)/l->cols; /* rows used by current buf. */ + int rpos = (plen+l->oldpos+l->cols)/l->cols; /* cursor relative row. */ + int rpos2; /* rpos after refresh. */ + int col; /* colum position, zero-based. */ + int old_rows = l->oldrows; + int fd = l->ofd, j; + std::string ab; + l->oldrows = rows; + + /* First step: clear all the lines used before. To do so start by + * going to the last row. */ + if (flags & REFRESH_CLEAN) { + if (old_rows-rpos > 0) { + lndebug("go down %d", old_rows-rpos); + snprintf(seq,64,"\x1b[%dB", old_rows-rpos); + ab.append(seq); + } + + /* Now for every row clear it, go up. */ + for (j = 0; j < old_rows-1; j++) { + lndebug("clear+up"); + snprintf(seq,64,"\r\x1b[0K\x1b[1A"); + ab.append(seq); + } + } + + if (flags & REFRESH_ALL) { + /* Clean the top line. */ + lndebug("clear"); + snprintf(seq,64,"\r\x1b[0K"); + ab.append(seq); + } + + if (flags & REFRESH_WRITE) { + /* Write the prompt and the current buffer content */ + ab.append(l->prompt); + if (maskmode == 1) { + for (unsigned int i = 0; i < l->len; ++i) { + ab.append("*"); + } + } else { + ab.append(l->buf, l->len); + } + + /* Show hits if any. */ + refreshShowHints(ab, l, plen); + + /* If we are at the very end of the screen with our prompt, we need to + * emit a newline and move the prompt to the first column. */ + if (l->pos && + l->pos == l->len && + (l->pos+plen) % l->cols == 0) + { + lndebug(""); + ab.append("\n"); + snprintf(seq,64,"\r"); + ab.append(seq); + rows++; + if (rows > (int)l->oldrows) l->oldrows = rows; + } + + /* Move cursor to right position. */ + rpos2 = (plen+l->pos+l->cols)/l->cols; /* Current cursor relative row */ + lndebug("rpos2 %d", rpos2); + + /* Go up till we reach the expected positon. */ + if (rows-rpos2 > 0) { + lndebug("go-up %d", rows-rpos2); + snprintf(seq,64,"\x1b[%dA", rows-rpos2); + ab.append(seq); + } + + /* Set column. */ + col = (plen+(int)l->pos) % (int)l->cols; + lndebug("set col %d", 1+col); + if (col) + snprintf(seq,64,"\r\x1b[%dC", col); + else + snprintf(seq,64,"\r"); + ab.append(seq); + } + + lndebug("\n"); + l->oldpos = l->pos; + (void) !write(fd, ab.c_str(), ab.size()); /* Can't recover from write error. */ +} + +/* Calls the two low level functions refreshSingleLine() or + * refreshMultiLine() according to the selected mode. */ +static void refreshLineWithFlags(struct linenoiseState *l, int flags) { + if (mlmode) + refreshMultiLine(l,flags); + else + refreshSingleLine(l,flags); +} + +/* Utility function to avoid specifying REFRESH_ALL all the times. */ +static void refreshLine(struct linenoiseState *l) { + refreshLineWithFlags(l,REFRESH_ALL); +} + +/* Hide the current line, when using the multiplexing API. */ +void linenoiseHide(struct linenoiseState *l) { + if (mlmode) + refreshMultiLine(l,REFRESH_CLEAN); + else + refreshSingleLine(l,REFRESH_CLEAN); +} + +/* Show the current line, when using the multiplexing API. */ +void linenoiseShow(struct linenoiseState *l) { + if (l->in_completion) { + refreshLineWithCompletion(l,NULL,REFRESH_WRITE); + } else { + refreshLineWithFlags(l,REFRESH_WRITE); + } +} + +/* Insert the character 'c' at cursor current position. + * + * On error writing to the terminal -1 is returned, otherwise 0. */ +static int linenoiseEditInsert(struct linenoiseState * l, char c) { + if (l->len < l->buflen) { + if (l->len == l->pos) { + l->buf[l->pos] = c; + l->pos++; + l->len++; + l->buf[l->len] = '\0'; + if ((!mlmode && l->plen+l->len < l->cols && !hintsCallback)) { + /* Avoid a full update of the line in the + * trivial case. */ + char d = (maskmode==1) ? '*' : c; + if (write(l->ofd,&d,1) == -1) return -1; + } else { + refreshLine(l); + } + } else { + memmove(l->buf+l->pos+1,l->buf+l->pos,l->len-l->pos); + l->buf[l->pos] = c; + l->len++; + l->pos++; + l->buf[l->len] = '\0'; + refreshLine(l); + } + } + return 0; +} + +/* Move cursor on the left. */ +static void linenoiseEditMoveLeft(struct linenoiseState * l) { + if (l->pos > 0) { + l->pos--; + refreshLine(l); + } +} + +/* Move cursor on the right. */ +static void linenoiseEditMoveRight(struct linenoiseState * l) { + if (l->pos != l->len) { + l->pos++; + refreshLine(l); + } +} + +/* Move cursor to the start of the line. */ +static void linenoiseEditMoveHome(struct linenoiseState * l) { + if (l->pos != 0) { + l->pos = 0; + refreshLine(l); + } +} + +/* Move cursor to the end of the line. */ +static void linenoiseEditMoveEnd(struct linenoiseState * l) { + if (l->pos != l->len) { + l->pos = l->len; + refreshLine(l); + } +} + +/* Substitute the currently edited line with the next or previous history + * entry as specified by 'dir'. */ +#define LINENOISE_HISTORY_NEXT 0 +#define LINENOISE_HISTORY_PREV 1 + +static void linenoiseEditHistoryNext(struct linenoiseState * l, int dir) { + if (history_len > 1) { + /* Update the current history entry before to + * overwrite it with the next one. */ + free(history[history_len - 1 - l->history_index]); + history[history_len - 1 - l->history_index] = strdup(l->buf); + /* Show the new entry */ + l->history_index += (dir == LINENOISE_HISTORY_PREV) ? 1 : -1; + if (l->history_index < 0) { + l->history_index = 0; + return; + } else if (l->history_index >= history_len) { + l->history_index = history_len-1; + return; + } + strncpy(l->buf,history[history_len - 1 - l->history_index],l->buflen); + l->buf[l->buflen-1] = '\0'; + l->len = l->pos = strlen(l->buf); + refreshLine(l); + } +} + +/* Delete the character at the right of the cursor without altering the cursor + * position. Basically this is what happens with the "Delete" keyboard key. */ +static void linenoiseEditDelete(struct linenoiseState * l) { + if (l->len > 0 && l->pos < l->len) { + memmove(l->buf+l->pos,l->buf+l->pos+1,l->len-l->pos-1); + l->len--; + l->buf[l->len] = '\0'; + refreshLine(l); + } +} + +/* Backspace implementation. */ +static void linenoiseEditBackspace(struct linenoiseState * l) { + if (l->pos > 0 && l->len > 0) { + memmove(l->buf+l->pos-1,l->buf+l->pos,l->len-l->pos); + l->pos--; + l->len--; + l->buf[l->len] = '\0'; + refreshLine(l); + } +} + +/* Delete the previosu word, maintaining the cursor at the start of the + * current word. */ +static void linenoiseEditDeletePrevWord(struct linenoiseState * l) { + size_t old_pos = l->pos; + size_t diff; + + while (l->pos > 0 && l->buf[l->pos-1] == ' ') + l->pos--; + while (l->pos > 0 && l->buf[l->pos-1] != ' ') + l->pos--; + diff = old_pos - l->pos; + memmove(l->buf+l->pos,l->buf+old_pos,l->len-old_pos+1); + l->len -= diff; + refreshLine(l); +} + +/* This function is part of the multiplexed API of Linenoise, that is used + * in order to implement the blocking variant of the API but can also be + * called by the user directly in an event driven program. It will: + * + * 1. Initialize the linenoise state passed by the user. + * 2. Put the terminal in RAW mode. + * 3. Show the prompt. + * 4. Return control to the user, that will have to call linenoiseEditFeed() + * each time there is some data arriving in the standard input. + * + * The user can also call linenoiseEditHide() and linenoiseEditShow() if it + * is required to show some input arriving asyncronously, without mixing + * it with the currently edited line. + * + * When linenoiseEditFeed() returns non-NULL, the user finished with the + * line editing session (pressed enter CTRL-D/C): in this case the caller + * needs to call linenoiseEditStop() to put back the terminal in normal + * mode. This will not destroy the buffer, as long as the linenoiseState + * is still valid in the context of the caller. + * + * The function returns 0 on success, or -1 if writing to standard output + * fails. If stdin_fd or stdout_fd are set to -1, the default is to use + * STDIN_FILENO and STDOUT_FILENO. + */ +int linenoiseEditStart(struct linenoiseState *l, int stdin_fd, int stdout_fd, char *buf, size_t buflen, const char *prompt) { + /* Populate the linenoise state that we pass to functions implementing + * specific editing functionalities. */ + l->in_completion = 0; + l->ifd = stdin_fd != -1 ? stdin_fd : STDIN_FILENO; + l->ofd = stdout_fd != -1 ? stdout_fd : STDOUT_FILENO; + l->buf = buf; + l->buflen = buflen; + l->prompt = prompt; + l->plen = strlen(prompt); + l->oldpos = l->pos = 0; + l->len = 0; + + /* Enter raw mode. */ + if (enableRawMode(l->ifd) == -1) return -1; + + l->cols = getColumns(stdin_fd, stdout_fd); + l->oldrows = 0; + l->history_index = 0; + + /* Buffer starts empty. */ + l->buf[0] = '\0'; + l->buflen--; /* Make sure there is always space for the nulterm */ + + /* If stdin is not a tty, stop here with the initialization. We + * will actually just read a line from standard input in blocking + * mode later, in linenoiseEditFeed(). */ + if (!isatty(l->ifd)) return 0; + + /* The latest history entry is always our current buffer, that + * initially is just an empty string. */ + linenoiseHistoryAdd(""); + + if (write(l->ofd,prompt,l->plen) == -1) return -1; + return 0; +} + +const char* linenoiseEditMore = "If you see this, you are misusing the API: when linenoiseEditFeed() is called, if it returns linenoiseEditMore the user is yet editing the line. See the README file for more information."; + +/* This function is part of the multiplexed API of linenoise, see the top + * comment on linenoiseEditStart() for more information. Call this function + * each time there is some data to read from the standard input file + * descriptor. In the case of blocking operations, this function can just be + * called in a loop, and block. + * + * The function returns linenoiseEditMore to signal that line editing is still + * in progress, that is, the user didn't yet pressed enter / CTRL-D. Otherwise + * the function returns the pointer to the heap-allocated buffer with the + * edited line, that the user should free with linenoiseFree(). + * + * On special conditions, NULL is returned and errno is populated: + * + * EAGAIN if the user pressed Ctrl-C + * ENOENT if the user pressed Ctrl-D + * + * Some other errno: I/O error. + */ +const char *linenoiseEditFeed(struct linenoiseState *l) { + /* Not a TTY, pass control to line reading without character + * count limits. */ + if (!isatty(l->ifd)) return linenoiseNoTTY(); + + char c; + int nread; + char seq[3]; + + nread = read(l->ifd,&c,1); + if (nread <= 0) return NULL; + + /* Only autocomplete when the callback is set. It returns < 0 when + * there was an error reading from fd. Otherwise it will return the + * character that should be handled next. */ + if ((l->in_completion || c == 9) && completionCallback != NULL) { + c = completeLine(l,c); + /* Read next character when 0 */ + if (c == 0) return linenoiseEditMore; + } + + switch(c) { + case ENTER: /* enter */ + history_len--; + free(history[history_len]); + if (mlmode) linenoiseEditMoveEnd(l); + if (hintsCallback) { + /* Force a refresh without hints to leave the previous + * line as the user typed it after a newline. */ + linenoiseHintsCallback *hc = hintsCallback; + hintsCallback = NULL; + refreshLine(l); + hintsCallback = hc; + } + return strdup(l->buf); + case CTRL_C: /* ctrl-c */ + errno = EAGAIN; + return NULL; + case BACKSPACE: /* backspace */ + case 8: /* ctrl-h */ + linenoiseEditBackspace(l); + break; + case CTRL_D: /* ctrl-d, remove char at right of cursor, or if the + line is empty, act as end-of-file. */ + if (l->len > 0) { + linenoiseEditDelete(l); + } else { + history_len--; + free(history[history_len]); + errno = ENOENT; + return NULL; + } + break; + case CTRL_T: /* ctrl-t, swaps current character with previous. */ + if (l->pos > 0 && l->pos < l->len) { + int aux = l->buf[l->pos-1]; + l->buf[l->pos-1] = l->buf[l->pos]; + l->buf[l->pos] = aux; + if (l->pos != l->len-1) l->pos++; + refreshLine(l); + } + break; + case CTRL_B: /* ctrl-b */ + linenoiseEditMoveLeft(l); + break; + case CTRL_F: /* ctrl-f */ + linenoiseEditMoveRight(l); + break; + case CTRL_P: /* ctrl-p */ + linenoiseEditHistoryNext(l, LINENOISE_HISTORY_PREV); + break; + case CTRL_N: /* ctrl-n */ + linenoiseEditHistoryNext(l, LINENOISE_HISTORY_NEXT); + break; + case ESC: /* escape sequence */ + /* Read the next two bytes representing the escape sequence. + * Use two calls to handle slow terminals returning the two + * chars at different times. */ + if (read(l->ifd,seq,1) == -1) break; + if (read(l->ifd,seq+1,1) == -1) break; + + /* ESC [ sequences. */ + if (seq[0] == '[') { + if (seq[1] >= '0' && seq[1] <= '9') { + /* Extended escape, read additional byte. */ + if (read(l->ifd,seq+2,1) == -1) break; + if (seq[2] == '~') { + switch(seq[1]) { + case '3': /* Delete key. */ + linenoiseEditDelete(l); + break; + } + } + } else { + switch(seq[1]) { + case 'A': /* Up */ + linenoiseEditHistoryNext(l, LINENOISE_HISTORY_PREV); + break; + case 'B': /* Down */ + linenoiseEditHistoryNext(l, LINENOISE_HISTORY_NEXT); + break; + case 'C': /* Right */ + linenoiseEditMoveRight(l); + break; + case 'D': /* Left */ + linenoiseEditMoveLeft(l); + break; + case 'H': /* Home */ + linenoiseEditMoveHome(l); + break; + case 'F': /* End*/ + linenoiseEditMoveEnd(l); + break; + } + } + } + + /* ESC O sequences. */ + else if (seq[0] == 'O') { + switch(seq[1]) { + case 'H': /* Home */ + linenoiseEditMoveHome(l); + break; + case 'F': /* End*/ + linenoiseEditMoveEnd(l); + break; + } + } + break; + default: + if (linenoiseEditInsert(l,c)) return NULL; + break; + case CTRL_U: /* Ctrl+u, delete the whole line. */ + l->buf[0] = '\0'; + l->pos = l->len = 0; + refreshLine(l); + break; + case CTRL_K: /* Ctrl+k, delete from current to end of line. */ + l->buf[l->pos] = '\0'; + l->len = l->pos; + refreshLine(l); + break; + case CTRL_A: /* Ctrl+a, go to the start of the line */ + linenoiseEditMoveHome(l); + break; + case CTRL_E: /* ctrl+e, go to the end of the line */ + linenoiseEditMoveEnd(l); + break; + case CTRL_L: /* ctrl+l, clear screen */ + linenoiseClearScreen(); + refreshLine(l); + break; + case CTRL_W: /* ctrl+w, delete previous word */ + linenoiseEditDeletePrevWord(l); + break; + } + return linenoiseEditMore; +} + +/* This is part of the multiplexed linenoise API. See linenoiseEditStart() + * for more information. This function is called when linenoiseEditFeed() + * returns something different than NULL. At this point the user input + * is in the buffer, and we can restore the terminal in normal mode. */ +void linenoiseEditStop(struct linenoiseState *l) { + if (!isatty(l->ifd)) return; + disableRawMode(l->ifd); + printf("\n"); +} + +/* This just implements a blocking loop for the multiplexed API. + * In many applications that are not event-drivern, we can just call + * the blocking linenoise API, wait for the user to complete the editing + * and return the buffer. */ +static const char *linenoiseBlockingEdit(int stdin_fd, int stdout_fd, char *buf, size_t buflen, const char *prompt) +{ + struct linenoiseState l; + + /* Editing without a buffer is invalid. */ + if (buflen == 0) { + errno = EINVAL; + return NULL; + } + + linenoiseEditStart(&l,stdin_fd,stdout_fd,buf,buflen,prompt); + const char *res; + while((res = linenoiseEditFeed(&l)) == linenoiseEditMore); + linenoiseEditStop(&l); + return res; +} + +/* This special mode is used by linenoise in order to print scan codes + * on screen for debugging / development purposes. It is implemented + * by the linenoise_example program using the --keycodes option. */ +void linenoisePrintKeyCodes(void) { + char quit[4]; + + printf("Linenoise key codes debugging mode.\n" + "Press keys to see scan codes. Type 'quit' at any time to exit.\n"); + if (enableRawMode(STDIN_FILENO) == -1) return; + memset(quit,' ',4); + while(1) { + char c; + int nread; + + nread = read(STDIN_FILENO,&c,1); + if (nread <= 0) continue; + memmove(quit,quit+1,sizeof(quit)-1); /* shift string to left. */ + quit[sizeof(quit)-1] = c; /* Insert current char on the right. */ + if (memcmp(quit,"quit",sizeof(quit)) == 0) break; + + printf("'%c' %02x (%d) (type quit to exit)\n", + isprint(c) ? c : '?', (int)c, (int)c); + printf("\r"); /* Go left edge manually, we are in raw mode. */ + fflush(stdout); + } + disableRawMode(STDIN_FILENO); +} + +/* This function is called when linenoise() is called with the standard + * input file descriptor not attached to a TTY. So for example when the + * program using linenoise is called in pipe or with a file redirected + * to its standard input. In this case, we want to be able to return the + * line regardless of its length (by default we are limited to 4k). */ +static char *linenoiseNoTTY(void) { + char *line = NULL; + size_t len = 0, maxlen = 0; + + while(1) { + if (len == maxlen) { + if (maxlen == 0) maxlen = 16; + maxlen *= 2; + char *oldval = line; + line = (char*) realloc(line,maxlen); + if (line == NULL) { + if (oldval) free(oldval); + return NULL; + } + } + int c = fgetc(stdin); + if (c == EOF || c == '\n') { + if (c == EOF && len == 0) { + free(line); + return NULL; + } else { + line[len] = '\0'; + return line; + } + } else { + line[len] = c; + len++; + } + } +} + +/* The high level function that is the main API of the linenoise library. + * This function checks if the terminal has basic capabilities, just checking + * for a blacklist of stupid terminals, and later either calls the line + * editing function or uses dummy fgets() so that you will be able to type + * something even in the most desperate of the conditions. */ +const char *linenoise(const char *prompt) { + char buf[LINENOISE_MAX_LINE]; + + if (!isatty(STDIN_FILENO)) { + /* Not a tty: read from file / pipe. In this mode we don't want any + * limit to the line size, so we call a function to handle that. */ + return linenoiseNoTTY(); + } else if (isUnsupportedTerm()) { + size_t len; + + printf("%s",prompt); + fflush(stdout); + if (fgets(buf,LINENOISE_MAX_LINE,stdin) == NULL) return NULL; + len = strlen(buf); + while(len && (buf[len-1] == '\n' || buf[len-1] == '\r')) { + len--; + buf[len] = '\0'; + } + return strdup(buf); + } else { + const char *retval = linenoiseBlockingEdit(STDIN_FILENO,STDOUT_FILENO,buf,LINENOISE_MAX_LINE,prompt); + return retval; + } +} + +/* This is just a wrapper the user may want to call in order to make sure + * the linenoise returned buffer is freed with the same allocator it was + * created with. Useful when the main program is using an alternative + * allocator. */ +void linenoiseFree(void *ptr) { + if (ptr == linenoiseEditMore) return; // Protect from API misuse. + free(ptr); +} + +/* ================================ History ================================= */ + +/* Free the history, but does not reset it. Only used when we have to + * exit() to avoid memory leaks are reported by valgrind & co. */ +static void freeHistory(void) { + if (history) { + int j; + + for (j = 0; j < history_len; j++) + free(history[j]); + free(history); + } +} + +/* At exit we'll try to fix the terminal to the initial conditions. */ +static void linenoiseAtExit(void) { + disableRawMode(STDIN_FILENO); + freeHistory(); +} + +/* This is the API call to add a new entry in the linenoise history. + * It uses a fixed array of char pointers that are shifted (memmoved) + * when the history max length is reached in order to remove the older + * entry and make room for the new one, so it is not exactly suitable for huge + * histories, but will work well for a few hundred of entries. + * + * Using a circular buffer is smarter, but a bit more complex to handle. */ +int linenoiseHistoryAdd(const char *line) { + char *linecopy; + + if (history_max_len == 0) return 0; + + /* Initialization on first call. */ + if (history == NULL) { + history = (char**) malloc(sizeof(char*)*history_max_len); + if (history == NULL) return 0; + memset(history,0,(sizeof(char*)*history_max_len)); + } + + /* Don't add duplicated lines. */ + if (history_len && !strcmp(history[history_len-1], line)) return 0; + + /* Add an heap allocated copy of the line in the history. + * If we reached the max length, remove the older line. */ + linecopy = strdup(line); + if (!linecopy) return 0; + if (history_len == history_max_len) { + free(history[0]); + memmove(history,history+1,sizeof(char*)*(history_max_len-1)); + history_len--; + } + history[history_len] = linecopy; + history_len++; + return 1; +} + +/* Set the maximum length for the history. This function can be called even + * if there is already some history, the function will make sure to retain + * just the latest 'len' elements if the new history length value is smaller + * than the amount of items already inside the history. */ +int linenoiseHistorySetMaxLen(int len) { + char **new_ptr; + + if (len < 1) return 0; + if (history) { + int tocopy = history_len; + + new_ptr = (char**) malloc(sizeof(char*)*len); + if (new_ptr == NULL) return 0; + + /* If we can't copy everything, free the elements we'll not use. */ + if (len < tocopy) { + int j; + + for (j = 0; j < tocopy-len; j++) free(history[j]); + tocopy = len; + } + memset(new_ptr,0,sizeof(char*)*len); + memcpy(new_ptr,history+(history_len-tocopy), sizeof(char*)*tocopy); + free(history); + history = new_ptr; + } + history_max_len = len; + if (history_len > history_max_len) + history_len = history_max_len; + return 1; +} + +/* Save the history in the specified file. On success 0 is returned + * otherwise -1 is returned. */ +int linenoiseHistorySave(const char *filename) { + mode_t old_umask = umask(S_IXUSR|S_IRWXG|S_IRWXO); + File file; + file.open(filename, "w"); + umask(old_umask); + if (file.file == NULL) { + return -1; + } + chmod(filename,S_IRUSR|S_IWUSR); + for (int j = 0; j < history_len; ++j) { + fprintf(file.file, "%s\n", history[j]); + } + + return 0; +} + +/* Load the history from the specified file. If the file does not exist + * zero is returned and no operation is performed. + * + * If the file exists and the operation succeeded 0 is returned, otherwise + * on error -1 is returned. */ +int linenoiseHistoryLoad(const char *filename) { + File file; + file.open(filename, "r"); + char buf[LINENOISE_MAX_LINE]; + if (file.file == NULL) { + return -1; + } + + while (fgets(buf, LINENOISE_MAX_LINE, file.file) != NULL) { + char *p; + + p = strchr(buf,'\r'); + if (!p) p = strchr(buf,'\n'); + if (p) *p = '\0'; + linenoiseHistoryAdd(buf); + } + return 0; +} +#endif diff --git a/examples/run/linenoise.cpp/linenoise.h b/examples/run/linenoise.cpp/linenoise.h new file mode 100644 index 000000000..a14ec6c74 --- /dev/null +++ b/examples/run/linenoise.cpp/linenoise.h @@ -0,0 +1,128 @@ +/* linenoise.h -- VERSION 1.0 + * + * Guerrilla line editing library against the idea that a line editing lib + * needs to be 20,000 lines of C++ code. + * + * See linenoise.cpp for more information. + * + * ------------------------------------------------------------------------ + * + * Copyright (c) 2010-2023, Salvatore Sanfilippo + * Copyright (c) 2010-2013, Pieter Noordhuis + * Copyright (c) 2025, Eric Curtin + * + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are + * met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef __LINENOISE_H +#define __LINENOISE_H + +#ifdef __cplusplus +extern "C" { +#endif + +#include /* For size_t. */ +#include + +extern const char *linenoiseEditMore; + +/* The linenoiseState structure represents the state during line editing. + * We pass this state to functions implementing specific editing + * functionalities. */ +struct linenoiseState { + int in_completion; /* The user pressed TAB and we are now in completion + * mode, so input is handled by completeLine(). */ + size_t completion_idx; /* Index of next completion to propose. */ + int ifd; /* Terminal stdin file descriptor. */ + int ofd; /* Terminal stdout file descriptor. */ + char *buf; /* Edited line buffer. */ + size_t buflen; /* Edited line buffer size. */ + const char *prompt; /* Prompt to display. */ + size_t plen; /* Prompt length. */ + size_t pos; /* Current cursor position. */ + size_t oldpos; /* Previous refresh cursor position. */ + size_t len; /* Current edited line length. */ + size_t cols; /* Number of columns in terminal. */ + size_t oldrows; /* Rows used by last refrehsed line (multiline mode) */ + int history_index; /* The history index we are currently editing. */ +}; + +struct linenoiseCompletions { + size_t len = 0; + char ** cvec = nullptr; + bool to_free = true; + + ~linenoiseCompletions() { + if (!to_free) { + return; + } + + for (size_t i = 0; i < len; ++i) { + free(cvec[i]); + } + + free(cvec); + } +}; + +/* Non blocking API. */ +int linenoiseEditStart(struct linenoiseState *l, int stdin_fd, int stdout_fd, char *buf, size_t buflen, const char *prompt); +const char *linenoiseEditFeed(struct linenoiseState *l); +void linenoiseEditStop(struct linenoiseState *l); +void linenoiseHide(struct linenoiseState *l); +void linenoiseShow(struct linenoiseState *l); + +/* Blocking API. */ +const char *linenoise(const char *prompt); +void linenoiseFree(void *ptr); + +/* Completion API. */ +typedef void(linenoiseCompletionCallback)(const char *, linenoiseCompletions *); +typedef const char*(linenoiseHintsCallback)(const char *, int *color, int *bold); +typedef void(linenoiseFreeHintsCallback)(const char *); +void linenoiseSetCompletionCallback(linenoiseCompletionCallback *); +void linenoiseSetHintsCallback(linenoiseHintsCallback *); +void linenoiseSetFreeHintsCallback(linenoiseFreeHintsCallback *); +void linenoiseAddCompletion(linenoiseCompletions *, const char *); + +/* History API. */ +int linenoiseHistoryAdd(const char *line); +int linenoiseHistorySetMaxLen(int len); +int linenoiseHistorySave(const char *filename); +int linenoiseHistoryLoad(const char *filename); + +/* Other utilities. */ +void linenoiseClearScreen(void); +void linenoiseSetMultiLine(int ml); +void linenoisePrintKeyCodes(void); +void linenoiseMaskModeEnable(void); +void linenoiseMaskModeDisable(void); + +#ifdef __cplusplus +} +#endif + +#endif /* __LINENOISE_H */ diff --git a/examples/run/run.cpp b/examples/run/run.cpp index 75b817272..92a49eb74 100644 --- a/examples/run/run.cpp +++ b/examples/run/run.cpp @@ -11,19 +11,31 @@ # include #endif +#include + #include #include #include #include #include #include +#include #include #include #include #include "common.h" #include "json.hpp" +#include "linenoise.cpp/linenoise.h" #include "llama-cpp.h" +#include "chat-template.hpp" + +#if defined(__unix__) || (defined(__APPLE__) && defined(__MACH__)) || defined(_WIN32) +[[noreturn]] static void sigint_handler(int) { + printf("\n\033[0m"); + exit(0); // not ideal, but it's the only way to guarantee exit in all cases +} +#endif GGML_ATTRIBUTE_FORMAT(1, 2) static std::string fmt(const char * fmt, ...) { @@ -83,6 +95,7 @@ class Opt { } ctx_params.n_batch = context_size >= 0 ? context_size : context_size_default; + ctx_params.n_ctx = ctx_params.n_batch; model_params.n_gpu_layers = ngl >= 0 ? ngl : ngl_default; temperature = temperature >= 0 ? temperature : temperature_default; @@ -93,6 +106,7 @@ class Opt { llama_model_params model_params; std::string model_; std::string user; + bool use_jinja = false; int context_size = -1, ngl = -1; float temperature = -1; bool verbose = false; @@ -133,7 +147,8 @@ class Opt { if (handle_option_with_value(argc, argv, i, context_size) == 1) { return 1; } - } else if (options_parsing && (strcmp(argv[i], "-n") == 0 || strcmp(argv[i], "--ngl") == 0)) { + } else if (options_parsing && + (strcmp(argv[i], "-n") == 0 || strcmp(argv[i], "-ngl") == 0 || strcmp(argv[i], "--ngl") == 0)) { if (handle_option_with_value(argc, argv, i, ngl) == 1) { return 1; } @@ -144,6 +159,8 @@ class Opt { } else if (options_parsing && (parse_flag(argv, i, "-v", "--verbose") || parse_flag(argv, i, "-v", "--log-verbose"))) { verbose = true; + } else if (options_parsing && strcmp(argv[i], "--jinja") == 0) { + use_jinja = true; } else if (options_parsing && parse_flag(argv, i, "-h", "--help")) { help = true; return 0; @@ -178,7 +195,7 @@ class Opt { "Options:\n" " -c, --context-size \n" " Context size (default: %d)\n" - " -n, --ngl \n" + " -n, -ngl, --ngl \n" " Number of GPU layers (default: %d)\n" " --temp \n" " Temperature (default: %.1f)\n" @@ -526,7 +543,7 @@ class LlamaData { llama_sampler_ptr sampler; llama_context_ptr context; std::vector messages; - std::vector msg_strs; + std::list msg_strs; std::vector fmtted; int init(Opt & opt) { @@ -618,20 +635,20 @@ class LlamaData { return path.substr(pos + 1); } - int remove_proto(std::string & model_) { - const std::string::size_type pos = model_.find("://"); + int rm_until_substring(std::string & model_, const std::string & substring) { + const std::string::size_type pos = model_.find(substring); if (pos == std::string::npos) { return 1; } - model_ = model_.substr(pos + 3); // Skip past "://" + model_ = model_.substr(pos + substring.size()); // Skip past the substring return 0; } int resolve_model(std::string & model_) { int ret = 0; if (string_starts_with(model_, "file://") || std::filesystem::exists(model_)) { - remove_proto(model_); + rm_until_substring(model_, "://"); return ret; } @@ -640,13 +657,16 @@ class LlamaData { const std::vector headers = { "--header", "Accept: application/vnd.docker.distribution.manifest.v2+json" }; if (string_starts_with(model_, "hf://") || string_starts_with(model_, "huggingface://")) { - remove_proto(model_); + rm_until_substring(model_, "://"); + ret = huggingface_dl(model_, headers, bn); + } else if (string_starts_with(model_, "hf.co/")) { + rm_until_substring(model_, "hf.co/"); ret = huggingface_dl(model_, headers, bn); } else if (string_starts_with(model_, "ollama://")) { - remove_proto(model_); + rm_until_substring(model_, "://"); ret = ollama_dl(model_, headers, bn); } else if (string_starts_with(model_, "https://")) { - download(model_, headers, bn, true); + ret = download(model_, headers, bn, true); } else { ret = ollama_dl(model_, headers, bn); } @@ -664,7 +684,7 @@ class LlamaData { "\r%*s" "\rLoading model", get_terminal_width(), " "); - llama_model_ptr model(llama_load_model_from_file(opt.model_.c_str(), opt.model_params)); + llama_model_ptr model(llama_model_load_from_file(opt.model_.c_str(), opt.model_params)); if (!model) { printe("%s: error: unable to load model from file: %s\n", __func__, opt.model_.c_str()); } @@ -675,7 +695,7 @@ class LlamaData { // Initializes the context with the specified parameters llama_context_ptr initialize_context(const llama_model_ptr & model, const Opt & opt) { - llama_context_ptr context(llama_new_context_with_model(model.get(), opt.ctx_params)); + llama_context_ptr context(llama_init_from_model(model.get(), opt.ctx_params)); if (!context) { printe("%s: error: failed to create the llama_context\n", __func__); } @@ -701,13 +721,31 @@ static void add_message(const char * role, const std::string & text, LlamaData & } // Function to apply the chat template and resize `formatted` if needed -static int apply_chat_template(LlamaData & llama_data, const bool append) { +static int apply_chat_template(const common_chat_template & tmpl, LlamaData & llama_data, const bool append, bool use_jinja) { + if (use_jinja) { + json messages = json::array(); + for (const auto & msg : llama_data.messages) { + messages.push_back({ + {"role", msg.role}, + {"content", msg.content}, + }); + } + try { + auto result = tmpl.apply(messages, /* tools= */ json(), append); + llama_data.fmtted.resize(result.size() + 1); + memcpy(llama_data.fmtted.data(), result.c_str(), result.size() + 1); + return result.size(); + } catch (const std::exception & e) { + printe("failed to render the chat template: %s\n", e.what()); + return -1; + } + } int result = llama_chat_apply_template( - llama_data.model.get(), nullptr, llama_data.messages.data(), llama_data.messages.size(), append, + tmpl.source().c_str(), llama_data.messages.data(), llama_data.messages.size(), append, append ? llama_data.fmtted.data() : nullptr, append ? llama_data.fmtted.size() : 0); if (append && result > static_cast(llama_data.fmtted.size())) { llama_data.fmtted.resize(result); - result = llama_chat_apply_template(llama_data.model.get(), nullptr, llama_data.messages.data(), + result = llama_chat_apply_template(tmpl.source().c_str(), llama_data.messages.data(), llama_data.messages.size(), append, llama_data.fmtted.data(), llama_data.fmtted.size()); } @@ -716,11 +754,13 @@ static int apply_chat_template(LlamaData & llama_data, const bool append) { } // Function to tokenize the prompt -static int tokenize_prompt(const llama_model_ptr & model, const std::string & prompt, - std::vector & prompt_tokens) { - const int n_prompt_tokens = -llama_tokenize(model.get(), prompt.c_str(), prompt.size(), NULL, 0, true, true); +static int tokenize_prompt(const llama_vocab * vocab, const std::string & prompt, + std::vector & prompt_tokens, const LlamaData & llama_data) { + const bool is_first = llama_get_kv_cache_used_cells(llama_data.context.get()) == 0; + + const int n_prompt_tokens = -llama_tokenize(vocab, prompt.c_str(), prompt.size(), NULL, 0, is_first, true); prompt_tokens.resize(n_prompt_tokens); - if (llama_tokenize(model.get(), prompt.c_str(), prompt.size(), prompt_tokens.data(), prompt_tokens.size(), true, + if (llama_tokenize(vocab, prompt.c_str(), prompt.size(), prompt_tokens.data(), prompt_tokens.size(), is_first, true) < 0) { printe("failed to tokenize the prompt\n"); return -1; @@ -743,9 +783,9 @@ static int check_context_size(const llama_context_ptr & ctx, const llama_batch & } // convert the token to a string -static int convert_token_to_string(const llama_model_ptr & model, const llama_token token_id, std::string & piece) { +static int convert_token_to_string(const llama_vocab * vocab, const llama_token token_id, std::string & piece) { char buf[256]; - int n = llama_token_to_piece(model.get(), token_id, buf, sizeof(buf), 0, true); + int n = llama_token_to_piece(vocab, token_id, buf, sizeof(buf), 0, true); if (n < 0) { printe("failed to convert token to piece\n"); return 1; @@ -763,8 +803,10 @@ static void print_word_and_concatenate_to_response(const std::string & piece, st // helper function to evaluate a prompt and generate a response static int generate(LlamaData & llama_data, const std::string & prompt, std::string & response) { + const llama_vocab * vocab = llama_model_get_vocab(llama_data.model.get()); + std::vector tokens; - if (tokenize_prompt(llama_data.model, prompt, tokens) < 0) { + if (tokenize_prompt(vocab, prompt, tokens, llama_data) < 0) { return 1; } @@ -780,12 +822,12 @@ static int generate(LlamaData & llama_data, const std::string & prompt, std::str // sample the next token, check is it an end of generation? new_token_id = llama_sampler_sample(llama_data.sampler.get(), llama_data.context.get(), -1); - if (llama_token_is_eog(llama_data.model.get(), new_token_id)) { + if (llama_vocab_is_eog(vocab, new_token_id)) { break; } std::string piece; - if (convert_token_to_string(llama_data.model, new_token_id, piece)) { + if (convert_token_to_string(vocab, new_token_id, piece)) { return 1; } @@ -795,12 +837,45 @@ static int generate(LlamaData & llama_data, const std::string & prompt, std::str batch = llama_batch_get_one(&new_token_id, 1); } + printf("\033[0m"); return 0; } -static int read_user_input(std::string & user) { - std::getline(std::cin, user); - return user.empty(); // Should have data in happy path +static int read_user_input(std::string & user_input) { + static const char * prompt_prefix = "> "; +#ifdef WIN32 + printf( + "\r%*s" + "\r\033[0m%s", + get_terminal_width(), " ", prompt_prefix); + + std::getline(std::cin, user_input); + if (std::cin.eof()) { + printf("\n"); + return 1; + } +#else + std::unique_ptr line(const_cast(linenoise(prompt_prefix)), free); + if (!line) { + return 1; + } + + user_input = line.get(); +#endif + + if (user_input == "/bye") { + return 1; + } + + if (user_input.empty()) { + return 2; + } + +#ifndef WIN32 + linenoiseHistoryAdd(line.get()); +#endif + + return 0; // Should have data in happy path } // Function to generate a response based on the prompt @@ -822,8 +897,8 @@ static int generate_response(LlamaData & llama_data, const std::string & prompt, } // Helper function to apply the chat template and handle errors -static int apply_chat_template_with_error_handling(LlamaData & llama_data, const bool append, int & output_length) { - const int new_len = apply_chat_template(llama_data, append); +static int apply_chat_template_with_error_handling(const common_chat_template & tmpl, LlamaData & llama_data, const bool append, int & output_length, bool use_jinja) { + const int new_len = apply_chat_template(tmpl, llama_data, append, use_jinja); if (new_len < 0) { printe("failed to apply the chat template\n"); return -1; @@ -840,10 +915,6 @@ static int handle_user_input(std::string & user_input, const std::string & user) return 0; // No need for interactive input } - printf( - "\r%*s" - "\r\033[32m> \033[0m", - get_terminal_width(), " "); return read_user_input(user_input); // Returns true if input ends the loop } @@ -867,20 +938,41 @@ static bool is_stdout_a_terminal() { #endif } -// Function to tokenize the prompt -static int chat_loop(LlamaData & llama_data, const std::string & user) { +// Function to handle user input +static int get_user_input(std::string & user_input, const std::string & user) { + while (true) { + const int ret = handle_user_input(user_input, user); + if (ret == 1) { + return 1; + } + + if (ret == 2) { + continue; + } + + break; + } + + return 0; +} + +// Main chat loop function +static int chat_loop(LlamaData & llama_data, const std::string & user, bool use_jinja) { int prev_len = 0; llama_data.fmtted.resize(llama_n_ctx(llama_data.context.get())); + auto chat_templates = common_chat_templates_from_model(llama_data.model.get(), ""); + GGML_ASSERT(chat_templates.template_default); static const bool stdout_a_terminal = is_stdout_a_terminal(); while (true) { // Get user input std::string user_input; - while (handle_user_input(user_input, user)) { + if (get_user_input(user_input, user) == 1) { + return 0; } add_message("user", user.empty() ? user_input : user, llama_data); int new_len; - if (apply_chat_template_with_error_handling(llama_data, true, new_len) < 0) { + if (apply_chat_template_with_error_handling(*chat_templates.template_default, llama_data, true, new_len, use_jinja) < 0) { return 1; } @@ -895,7 +987,7 @@ static int chat_loop(LlamaData & llama_data, const std::string & user) { } add_message("assistant", response, llama_data); - if (apply_chat_template_with_error_handling(llama_data, false, prev_len) < 0) { + if (apply_chat_template_with_error_handling(*chat_templates.template_default, llama_data, false, prev_len, use_jinja) < 0) { return 1; } } @@ -916,7 +1008,23 @@ static std::string read_pipe_data() { return result.str(); } +static void ctrl_c_handling() { +#if defined(__unix__) || (defined(__APPLE__) && defined(__MACH__)) + struct sigaction sigint_action; + sigint_action.sa_handler = sigint_handler; + sigemptyset(&sigint_action.sa_mask); + sigint_action.sa_flags = 0; + sigaction(SIGINT, &sigint_action, NULL); +#elif defined(_WIN32) + auto console_ctrl_handler = +[](DWORD ctrl_type) -> BOOL { + return (ctrl_type == CTRL_C_EVENT) ? (sigint_handler(SIGINT), true) : false; + }; + SetConsoleCtrlHandler(reinterpret_cast(console_ctrl_handler), true); +#endif +} + int main(int argc, const char ** argv) { + ctrl_c_handling(); Opt opt; const int ret = opt.init(argc, argv); if (ret == 2) { @@ -939,7 +1047,7 @@ int main(int argc, const char ** argv) { return 1; } - if (chat_loop(llama_data, opt.user)) { + if (chat_loop(llama_data, opt.user, opt.use_jinja)) { return 1; } diff --git a/examples/save-load-state/save-load-state.cpp b/examples/save-load-state/save-load-state.cpp index cd03661cf..cf7cbd815 100644 --- a/examples/save-load-state/save-load-state.cpp +++ b/examples/save-load-state/save-load-state.cpp @@ -97,7 +97,7 @@ int main(int argc, char ** argv) { printf("\n\n"); // make new context - llama_context * ctx2 = llama_new_context_with_model(model, common_context_params_to_llama(params)); + llama_context * ctx2 = llama_init_from_model(model, common_context_params_to_llama(params)); llama_sampler * smpl2 = llama_sampler_chain_init(sparams); @@ -154,7 +154,7 @@ int main(int argc, char ** argv) { } // make new context - llama_context * ctx3 = llama_new_context_with_model(model, common_context_params_to_llama(params)); + llama_context * ctx3 = llama_init_from_model(model, common_context_params_to_llama(params)); llama_sampler * smpl3 = llama_sampler_chain_init(sparams); diff --git a/examples/server/README.md b/examples/server/README.md index 3ce16945a..5022de672 100644 --- a/examples/server/README.md +++ b/examples/server/README.md @@ -45,10 +45,7 @@ The project is under active development, and we are [looking for feedback and co | `-ub, --ubatch-size N` | physical maximum batch size (default: 512)
(env: LLAMA_ARG_UBATCH) | | `--keep N` | number of tokens to keep from the initial prompt (default: 0, -1 = all) | | `-fa, --flash-attn` | enable Flash Attention (default: disabled)
(env: LLAMA_ARG_FLASH_ATTN) | -| `-p, --prompt PROMPT` | prompt to start generation with | | `--no-perf` | disable internal libllama performance timings (default: false)
(env: LLAMA_ARG_NO_PERF) | -| `-f, --file FNAME` | a file containing the prompt (default: none) | -| `-bf, --binary-file FNAME` | binary file containing the prompt (default: none) | | `-e, --escape` | process escapes sequences (\n, \r, \t, \', \", \\) (default: true) | | `--no-escape` | do not process escape sequences | | `--rope-scaling {none,linear,yarn}` | RoPE frequency scaling method, defaults to linear unless specified by the model
(env: LLAMA_ARG_ROPE_SCALING_TYPE) | @@ -129,7 +126,7 @@ The project is under active development, and we are [looking for feedback and co | `--grammar GRAMMAR` | BNF-like grammar to constrain generations (see samples in grammars/ dir) (default: '') | | `--grammar-file FNAME` | file to read grammar from | | `-j, --json-schema SCHEMA` | JSON schema to constrain generations (https://json-schema.org/), e.g. `{}` for any JSON object
For schemas w/ external $refs, use --grammar + example/json_schema_to_grammar.py instead | - +| `--jinja` | Enable experimental Jinja templating engine (needed for tool use) | **Example-specific params** diff --git a/examples/server/httplib.h b/examples/server/httplib.h index f360bd93e..c2f12dd2a 100644 --- a/examples/server/httplib.h +++ b/examples/server/httplib.h @@ -8,7 +8,7 @@ #ifndef CPPHTTPLIB_HTTPLIB_H #define CPPHTTPLIB_HTTPLIB_H -#define CPPHTTPLIB_VERSION "0.15.3" +#define CPPHTTPLIB_VERSION "0.18.5" /* * Configuration @@ -18,8 +18,12 @@ #define CPPHTTPLIB_KEEPALIVE_TIMEOUT_SECOND 5 #endif +#ifndef CPPHTTPLIB_KEEPALIVE_TIMEOUT_CHECK_INTERVAL_USECOND +#define CPPHTTPLIB_KEEPALIVE_TIMEOUT_CHECK_INTERVAL_USECOND 10000 +#endif + #ifndef CPPHTTPLIB_KEEPALIVE_MAX_COUNT -#define CPPHTTPLIB_KEEPALIVE_MAX_COUNT 5 +#define CPPHTTPLIB_KEEPALIVE_MAX_COUNT 100 #endif #ifndef CPPHTTPLIB_CONNECTION_TIMEOUT_SECOND @@ -30,20 +34,36 @@ #define CPPHTTPLIB_CONNECTION_TIMEOUT_USECOND 0 #endif -#ifndef CPPHTTPLIB_READ_TIMEOUT_SECOND -#define CPPHTTPLIB_READ_TIMEOUT_SECOND 5 +#ifndef CPPHTTPLIB_SERVER_READ_TIMEOUT_SECOND +#define CPPHTTPLIB_SERVER_READ_TIMEOUT_SECOND 5 #endif -#ifndef CPPHTTPLIB_READ_TIMEOUT_USECOND -#define CPPHTTPLIB_READ_TIMEOUT_USECOND 0 +#ifndef CPPHTTPLIB_SERVER_READ_TIMEOUT_USECOND +#define CPPHTTPLIB_SERVER_READ_TIMEOUT_USECOND 0 #endif -#ifndef CPPHTTPLIB_WRITE_TIMEOUT_SECOND -#define CPPHTTPLIB_WRITE_TIMEOUT_SECOND 5 +#ifndef CPPHTTPLIB_SERVER_WRITE_TIMEOUT_SECOND +#define CPPHTTPLIB_SERVER_WRITE_TIMEOUT_SECOND 5 #endif -#ifndef CPPHTTPLIB_WRITE_TIMEOUT_USECOND -#define CPPHTTPLIB_WRITE_TIMEOUT_USECOND 0 +#ifndef CPPHTTPLIB_SERVER_WRITE_TIMEOUT_USECOND +#define CPPHTTPLIB_SERVER_WRITE_TIMEOUT_USECOND 0 +#endif + +#ifndef CPPHTTPLIB_CLIENT_READ_TIMEOUT_SECOND +#define CPPHTTPLIB_CLIENT_READ_TIMEOUT_SECOND 300 +#endif + +#ifndef CPPHTTPLIB_CLIENT_READ_TIMEOUT_USECOND +#define CPPHTTPLIB_CLIENT_READ_TIMEOUT_USECOND 0 +#endif + +#ifndef CPPHTTPLIB_CLIENT_WRITE_TIMEOUT_SECOND +#define CPPHTTPLIB_CLIENT_WRITE_TIMEOUT_SECOND 5 +#endif + +#ifndef CPPHTTPLIB_CLIENT_WRITE_TIMEOUT_USECOND +#define CPPHTTPLIB_CLIENT_WRITE_TIMEOUT_USECOND 0 #endif #ifndef CPPHTTPLIB_IDLE_INTERVAL_SECOND @@ -90,8 +110,12 @@ #define CPPHTTPLIB_TCP_NODELAY false #endif +#ifndef CPPHTTPLIB_IPV6_V6ONLY +#define CPPHTTPLIB_IPV6_V6ONLY false +#endif + #ifndef CPPHTTPLIB_RECV_BUFSIZ -#define CPPHTTPLIB_RECV_BUFSIZ size_t(4096u) +#define CPPHTTPLIB_RECV_BUFSIZ size_t(16384u) #endif #ifndef CPPHTTPLIB_COMPRESSION_BUFSIZ @@ -145,11 +169,11 @@ using ssize_t = long; #endif // _MSC_VER #ifndef S_ISREG -#define S_ISREG(m) (((m)&S_IFREG) == S_IFREG) +#define S_ISREG(m) (((m) & S_IFREG) == S_IFREG) #endif // S_ISREG #ifndef S_ISDIR -#define S_ISDIR(m) (((m)&S_IFDIR) == S_IFDIR) +#define S_ISDIR(m) (((m) & S_IFDIR) == S_IFDIR) #endif // S_ISDIR #ifndef NOMINMAX @@ -269,7 +293,12 @@ using socket_t = int; #include #include -#if OPENSSL_VERSION_NUMBER < 0x30000000L +#if defined(OPENSSL_IS_BORINGSSL) || defined(LIBRESSL_VERSION_NUMBER) +#if OPENSSL_VERSION_NUMBER < 0x1010107f +#error Please use OpenSSL or a current version of BoringSSL +#endif +#define SSL_get1_peer_certificate SSL_get_peer_certificate +#elif OPENSSL_VERSION_NUMBER < 0x30000000L #error Sorry, OpenSSL versions prior to 3.0.0 are not supported #endif @@ -312,16 +341,63 @@ make_unique(std::size_t n) { return std::unique_ptr(new RT[n]); } -struct ci { - bool operator()(const std::string &s1, const std::string &s2) const { - return std::lexicographical_compare(s1.begin(), s1.end(), s2.begin(), - s2.end(), - [](unsigned char c1, unsigned char c2) { - return ::tolower(c1) < ::tolower(c2); - }); +namespace case_ignore { + +inline unsigned char to_lower(int c) { + const static unsigned char table[256] = { + 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, + 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, + 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, + 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, + 60, 61, 62, 63, 64, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, + 107, 108, 109, 110, 111, 112, 113, 114, 115, 116, 117, 118, 119, 120, 121, + 122, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, + 105, 106, 107, 108, 109, 110, 111, 112, 113, 114, 115, 116, 117, 118, 119, + 120, 121, 122, 123, 124, 125, 126, 127, 128, 129, 130, 131, 132, 133, 134, + 135, 136, 137, 138, 139, 140, 141, 142, 143, 144, 145, 146, 147, 148, 149, + 150, 151, 152, 153, 154, 155, 156, 157, 158, 159, 160, 161, 162, 163, 164, + 165, 166, 167, 168, 169, 170, 171, 172, 173, 174, 175, 176, 177, 178, 179, + 180, 181, 182, 183, 184, 185, 186, 187, 188, 189, 190, 191, 224, 225, 226, + 227, 228, 229, 230, 231, 232, 233, 234, 235, 236, 237, 238, 239, 240, 241, + 242, 243, 244, 245, 246, 215, 248, 249, 250, 251, 252, 253, 254, 223, 224, + 225, 226, 227, 228, 229, 230, 231, 232, 233, 234, 235, 236, 237, 238, 239, + 240, 241, 242, 243, 244, 245, 246, 247, 248, 249, 250, 251, 252, 253, 254, + 255, + }; + return table[(unsigned char)(char)c]; +} + +inline bool equal(const std::string &a, const std::string &b) { + return a.size() == b.size() && + std::equal(a.begin(), a.end(), b.begin(), [](char ca, char cb) { + return to_lower(ca) == to_lower(cb); + }); +} + +struct equal_to { + bool operator()(const std::string &a, const std::string &b) const { + return equal(a, b); } }; +struct hash { + size_t operator()(const std::string &key) const { + return hash_core(key.data(), key.size(), 0); + } + + size_t hash_core(const char *s, size_t l, size_t h) const { + return (l == 0) ? h + : hash_core(s + 1, l - 1, + // Unsets the 6 high bits of h, therefore no + // overflow happens + (((std::numeric_limits::max)() >> 6) & + h * 33) ^ + static_cast(to_lower(*s))); + } +}; + +} // namespace case_ignore + // This is based on // "http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2014/n4189". @@ -427,7 +503,9 @@ enum StatusCode { NetworkAuthenticationRequired_511 = 511, }; -using Headers = std::multimap; +using Headers = + std::unordered_multimap; using Params = std::multimap; using Match = std::smatch; @@ -534,6 +612,7 @@ using Ranges = std::vector; struct Request { std::string method; std::string path; + Params params; Headers headers; std::string body; @@ -545,11 +624,11 @@ struct Request { // for server std::string version; std::string target; - Params params; MultipartFormDataMap files; Ranges ranges; Match matches; std::unordered_map path_params; + std::function is_connection_closed = []() { return true; }; // for client ResponseHandler response_handler; @@ -560,8 +639,10 @@ struct Request { #endif bool has_header(const std::string &key) const; - std::string get_header_value(const std::string &key, size_t id = 0) const; - uint64_t get_header_value_u64(const std::string &key, size_t id = 0) const; + std::string get_header_value(const std::string &key, const char *def = "", + size_t id = 0) const; + uint64_t get_header_value_u64(const std::string &key, uint64_t def = 0, + size_t id = 0) const; size_t get_header_value_count(const std::string &key) const; void set_header(const std::string &key, const std::string &val); @@ -592,8 +673,10 @@ struct Response { std::string location; // Redirect location bool has_header(const std::string &key) const; - std::string get_header_value(const std::string &key, size_t id = 0) const; - uint64_t get_header_value_u64(const std::string &key, size_t id = 0) const; + std::string get_header_value(const std::string &key, const char *def = "", + size_t id = 0) const; + uint64_t get_header_value_u64(const std::string &key, uint64_t def = 0, + size_t id = 0) const; size_t get_header_value_count(const std::string &key) const; void set_header(const std::string &key, const std::string &val); @@ -614,6 +697,10 @@ struct Response { const std::string &content_type, ContentProviderWithoutLength provider, ContentProviderResourceReleaser resource_releaser = nullptr); + void set_file_content(const std::string &path, + const std::string &content_type); + void set_file_content(const std::string &path); + Response() = default; Response(const Response &) = default; Response &operator=(const Response &) = default; @@ -631,6 +718,8 @@ struct Response { ContentProviderResourceReleaser content_provider_resource_releaser_; bool is_chunked_content_provider_ = false; bool content_provider_success_ = false; + std::string file_content_path_; + std::string file_content_content_type_; }; class Stream { @@ -646,8 +735,6 @@ public: virtual void get_local_ip_and_port(std::string &ip, int &port) const = 0; virtual socket_t socket() const = 0; - template - ssize_t write_format(const char *fmt, const Args &...args); ssize_t write(const char *ptr); ssize_t write(const std::string &s); }; @@ -719,13 +806,18 @@ private: if (pool_.shutdown_ && pool_.jobs_.empty()) { break; } - fn = std::move(pool_.jobs_.front()); + fn = pool_.jobs_.front(); pool_.jobs_.pop_front(); } assert(true == static_cast(fn)); fn(); } + +#if defined(CPPHTTPLIB_OPENSSL_SUPPORT) && !defined(OPENSSL_IS_BORINGSSL) && \ + !defined(LIBRESSL_VERSION_NUMBER) + OPENSSL_thread_stop(); +#endif } ThreadPool &pool_; @@ -787,7 +879,6 @@ public: bool match(Request &request) const override; private: - static constexpr char marker = ':'; // Treat segment separators as the end of path parameter capture // Does not need to handle query parameters as they are parsed before path // matching @@ -871,8 +962,13 @@ public: Server &set_default_file_mimetype(const std::string &mime); Server &set_file_request_handler(Handler handler); - Server &set_error_handler(HandlerWithResponse handler); - Server &set_error_handler(Handler handler); + template + Server &set_error_handler(ErrorHandlerFunc &&handler) { + return set_error_handler_core( + std::forward(handler), + std::is_convertible{}); + } + Server &set_exception_handler(ExceptionHandler handler); Server &set_pre_routing_handler(HandlerWithResponse handler); Server &set_post_routing_handler(Handler handler); @@ -882,6 +978,7 @@ public: Server &set_address_family(int family); Server &set_tcp_nodelay(bool on); + Server &set_ipv6_v6only(bool on); Server &set_socket_options(SocketOptions socket_options); Server &set_default_headers(Headers headers); @@ -914,21 +1011,24 @@ public: bool is_running() const; void wait_until_ready() const; void stop(); + void decommission(); std::function new_task_queue; protected: - bool process_request(Stream &strm, bool close_connection, + bool process_request(Stream &strm, const std::string &remote_addr, + int remote_port, const std::string &local_addr, + int local_port, bool close_connection, bool &connection_closed, const std::function &setup_request); std::atomic svr_sock_{INVALID_SOCKET}; size_t keep_alive_max_count_ = CPPHTTPLIB_KEEPALIVE_MAX_COUNT; time_t keep_alive_timeout_sec_ = CPPHTTPLIB_KEEPALIVE_TIMEOUT_SECOND; - time_t read_timeout_sec_ = CPPHTTPLIB_READ_TIMEOUT_SECOND; - time_t read_timeout_usec_ = CPPHTTPLIB_READ_TIMEOUT_USECOND; - time_t write_timeout_sec_ = CPPHTTPLIB_WRITE_TIMEOUT_SECOND; - time_t write_timeout_usec_ = CPPHTTPLIB_WRITE_TIMEOUT_USECOND; + time_t read_timeout_sec_ = CPPHTTPLIB_SERVER_READ_TIMEOUT_SECOND; + time_t read_timeout_usec_ = CPPHTTPLIB_SERVER_READ_TIMEOUT_USECOND; + time_t write_timeout_sec_ = CPPHTTPLIB_SERVER_WRITE_TIMEOUT_SECOND; + time_t write_timeout_usec_ = CPPHTTPLIB_SERVER_WRITE_TIMEOUT_USECOND; time_t idle_interval_sec_ = CPPHTTPLIB_IDLE_INTERVAL_SECOND; time_t idle_interval_usec_ = CPPHTTPLIB_IDLE_INTERVAL_USECOND; size_t payload_max_length_ = CPPHTTPLIB_PAYLOAD_MAX_LENGTH; @@ -943,6 +1043,9 @@ private: static std::unique_ptr make_matcher(const std::string &pattern); + Server &set_error_handler_core(HandlerWithResponse handler, std::true_type); + Server &set_error_handler_core(Handler handler, std::false_type); + socket_t create_server_socket(const std::string &host, int port, int socket_flags, SocketOptions socket_options) const; @@ -985,7 +1088,7 @@ private: virtual bool process_and_close_socket(socket_t sock); std::atomic is_running_{false}; - std::atomic done_{false}; + std::atomic is_decommisioned{false}; struct MountPointEntry { std::string mount_point; @@ -1018,6 +1121,7 @@ private: int address_family_ = AF_UNSPEC; bool tcp_nodelay_ = CPPHTTPLIB_TCP_NODELAY; + bool ipv6_v6only_ = CPPHTTPLIB_IPV6_V6ONLY; SocketOptions socket_options_ = default_socket_options; Headers default_headers_; @@ -1037,6 +1141,7 @@ enum class Error { SSLConnection, SSLLoadingCerts, SSLServerVerification, + SSLServerHostnameVerification, UnsupportedMultipartBoundaryChars, Compression, ConnectionTimeout, @@ -1074,9 +1179,10 @@ public: // Request Headers bool has_request_header(const std::string &key) const; std::string get_request_header_value(const std::string &key, + const char *def = "", size_t id = 0) const; uint64_t get_request_header_value_u64(const std::string &key, - size_t id = 0) const; + uint64_t def = 0, size_t id = 0) const; size_t get_request_header_value_count(const std::string &key) const; private: @@ -1140,10 +1246,18 @@ public: const std::string &content_type); Result Post(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type); + Result Post(const std::string &path, const Headers &headers, const char *body, + size_t content_length, const std::string &content_type, + Progress progress); Result Post(const std::string &path, const std::string &body, const std::string &content_type); + Result Post(const std::string &path, const std::string &body, + const std::string &content_type, Progress progress); Result Post(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type); + Result Post(const std::string &path, const Headers &headers, + const std::string &body, const std::string &content_type, + Progress progress); Result Post(const std::string &path, size_t content_length, ContentProvider content_provider, const std::string &content_type); @@ -1159,6 +1273,8 @@ public: Result Post(const std::string &path, const Params ¶ms); Result Post(const std::string &path, const Headers &headers, const Params ¶ms); + Result Post(const std::string &path, const Headers &headers, + const Params ¶ms, Progress progress); Result Post(const std::string &path, const MultipartFormDataItems &items); Result Post(const std::string &path, const Headers &headers, const MultipartFormDataItems &items); @@ -1173,10 +1289,18 @@ public: const std::string &content_type); Result Put(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type); + Result Put(const std::string &path, const Headers &headers, const char *body, + size_t content_length, const std::string &content_type, + Progress progress); Result Put(const std::string &path, const std::string &body, const std::string &content_type); + Result Put(const std::string &path, const std::string &body, + const std::string &content_type, Progress progress); Result Put(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type); + Result Put(const std::string &path, const Headers &headers, + const std::string &body, const std::string &content_type, + Progress progress); Result Put(const std::string &path, size_t content_length, ContentProvider content_provider, const std::string &content_type); Result Put(const std::string &path, @@ -1191,6 +1315,8 @@ public: Result Put(const std::string &path, const Params ¶ms); Result Put(const std::string &path, const Headers &headers, const Params ¶ms); + Result Put(const std::string &path, const Headers &headers, + const Params ¶ms, Progress progress); Result Put(const std::string &path, const MultipartFormDataItems &items); Result Put(const std::string &path, const Headers &headers, const MultipartFormDataItems &items); @@ -1203,13 +1329,23 @@ public: Result Patch(const std::string &path); Result Patch(const std::string &path, const char *body, size_t content_length, const std::string &content_type); + Result Patch(const std::string &path, const char *body, size_t content_length, + const std::string &content_type, Progress progress); Result Patch(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type); + Result Patch(const std::string &path, const Headers &headers, + const char *body, size_t content_length, + const std::string &content_type, Progress progress); Result Patch(const std::string &path, const std::string &body, const std::string &content_type); + Result Patch(const std::string &path, const std::string &body, + const std::string &content_type, Progress progress); Result Patch(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type); + Result Patch(const std::string &path, const Headers &headers, + const std::string &body, const std::string &content_type, + Progress progress); Result Patch(const std::string &path, size_t content_length, ContentProvider content_provider, const std::string &content_type); @@ -1227,13 +1363,24 @@ public: Result Delete(const std::string &path, const Headers &headers); Result Delete(const std::string &path, const char *body, size_t content_length, const std::string &content_type); + Result Delete(const std::string &path, const char *body, + size_t content_length, const std::string &content_type, + Progress progress); Result Delete(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type); + Result Delete(const std::string &path, const Headers &headers, + const char *body, size_t content_length, + const std::string &content_type, Progress progress); Result Delete(const std::string &path, const std::string &body, const std::string &content_type); + Result Delete(const std::string &path, const std::string &body, + const std::string &content_type, Progress progress); Result Delete(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type); + Result Delete(const std::string &path, const Headers &headers, + const std::string &body, const std::string &content_type, + Progress progress); Result Options(const std::string &path); Result Options(const std::string &path, const Headers &headers); @@ -1258,6 +1405,7 @@ public: void set_address_family(int family); void set_tcp_nodelay(bool on); + void set_ipv6_v6only(bool on); void set_socket_options(SocketOptions socket_options); void set_connection_timeout(time_t sec, time_t usec = 0); @@ -1309,6 +1457,8 @@ public: #ifdef CPPHTTPLIB_OPENSSL_SUPPORT void enable_server_certificate_verification(bool enabled); + void enable_server_hostname_verification(bool enabled); + void set_server_certificate_verifier(std::function verifier); #endif void set_logger(Logger logger); @@ -1375,10 +1525,10 @@ protected: time_t connection_timeout_sec_ = CPPHTTPLIB_CONNECTION_TIMEOUT_SECOND; time_t connection_timeout_usec_ = CPPHTTPLIB_CONNECTION_TIMEOUT_USECOND; - time_t read_timeout_sec_ = CPPHTTPLIB_READ_TIMEOUT_SECOND; - time_t read_timeout_usec_ = CPPHTTPLIB_READ_TIMEOUT_USECOND; - time_t write_timeout_sec_ = CPPHTTPLIB_WRITE_TIMEOUT_SECOND; - time_t write_timeout_usec_ = CPPHTTPLIB_WRITE_TIMEOUT_USECOND; + time_t read_timeout_sec_ = CPPHTTPLIB_CLIENT_READ_TIMEOUT_SECOND; + time_t read_timeout_usec_ = CPPHTTPLIB_CLIENT_READ_TIMEOUT_USECOND; + time_t write_timeout_sec_ = CPPHTTPLIB_CLIENT_WRITE_TIMEOUT_SECOND; + time_t write_timeout_usec_ = CPPHTTPLIB_CLIENT_WRITE_TIMEOUT_USECOND; std::string basic_auth_username_; std::string basic_auth_password_; @@ -1395,6 +1545,7 @@ protected: int address_family_ = AF_UNSPEC; bool tcp_nodelay_ = CPPHTTPLIB_TCP_NODELAY; + bool ipv6_v6only_ = CPPHTTPLIB_IPV6_V6ONLY; SocketOptions socket_options_ = nullptr; bool compress_ = false; @@ -1422,6 +1573,8 @@ protected: #ifdef CPPHTTPLIB_OPENSSL_SUPPORT bool server_certificate_verification_ = true; + bool server_hostname_verification_ = true; + std::function server_certificate_verifier_; #endif Logger logger_; @@ -1430,6 +1583,9 @@ private: bool send_(Request &req, Response &res, Error &error); Result send_(Request &&req); +#ifdef CPPHTTPLIB_OPENSSL_SUPPORT + bool is_ssl_peer_could_be_closed(SSL *ssl) const; +#endif socket_t create_client_socket(Error &error) const; bool read_response_line(Stream &strm, const Request &req, Response &res) const; @@ -1448,7 +1604,7 @@ private: const Headers &headers, const char *body, size_t content_length, ContentProvider content_provider, ContentProviderWithoutLength content_provider_without_length, - const std::string &content_type); + const std::string &content_type, Progress progress); ContentProviderWithoutLength get_multipart_content_provider( const std::string &boundary, const MultipartFormDataItems &items, const MultipartFormDataProviderItems &provider_items) const; @@ -1477,6 +1633,7 @@ public: const std::string &client_key_path); Client(Client &&) = default; + Client &operator=(Client &&) = default; ~Client(); @@ -1523,10 +1680,18 @@ public: const std::string &content_type); Result Post(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type); + Result Post(const std::string &path, const Headers &headers, const char *body, + size_t content_length, const std::string &content_type, + Progress progress); Result Post(const std::string &path, const std::string &body, const std::string &content_type); + Result Post(const std::string &path, const std::string &body, + const std::string &content_type, Progress progress); Result Post(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type); + Result Post(const std::string &path, const Headers &headers, + const std::string &body, const std::string &content_type, + Progress progress); Result Post(const std::string &path, size_t content_length, ContentProvider content_provider, const std::string &content_type); @@ -1542,6 +1707,8 @@ public: Result Post(const std::string &path, const Params ¶ms); Result Post(const std::string &path, const Headers &headers, const Params ¶ms); + Result Post(const std::string &path, const Headers &headers, + const Params ¶ms, Progress progress); Result Post(const std::string &path, const MultipartFormDataItems &items); Result Post(const std::string &path, const Headers &headers, const MultipartFormDataItems &items); @@ -1556,10 +1723,18 @@ public: const std::string &content_type); Result Put(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type); + Result Put(const std::string &path, const Headers &headers, const char *body, + size_t content_length, const std::string &content_type, + Progress progress); Result Put(const std::string &path, const std::string &body, const std::string &content_type); + Result Put(const std::string &path, const std::string &body, + const std::string &content_type, Progress progress); Result Put(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type); + Result Put(const std::string &path, const Headers &headers, + const std::string &body, const std::string &content_type, + Progress progress); Result Put(const std::string &path, size_t content_length, ContentProvider content_provider, const std::string &content_type); Result Put(const std::string &path, @@ -1574,6 +1749,8 @@ public: Result Put(const std::string &path, const Params ¶ms); Result Put(const std::string &path, const Headers &headers, const Params ¶ms); + Result Put(const std::string &path, const Headers &headers, + const Params ¶ms, Progress progress); Result Put(const std::string &path, const MultipartFormDataItems &items); Result Put(const std::string &path, const Headers &headers, const MultipartFormDataItems &items); @@ -1586,13 +1763,23 @@ public: Result Patch(const std::string &path); Result Patch(const std::string &path, const char *body, size_t content_length, const std::string &content_type); + Result Patch(const std::string &path, const char *body, size_t content_length, + const std::string &content_type, Progress progress); Result Patch(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type); + Result Patch(const std::string &path, const Headers &headers, + const char *body, size_t content_length, + const std::string &content_type, Progress progress); Result Patch(const std::string &path, const std::string &body, const std::string &content_type); + Result Patch(const std::string &path, const std::string &body, + const std::string &content_type, Progress progress); Result Patch(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type); + Result Patch(const std::string &path, const Headers &headers, + const std::string &body, const std::string &content_type, + Progress progress); Result Patch(const std::string &path, size_t content_length, ContentProvider content_provider, const std::string &content_type); @@ -1610,13 +1797,24 @@ public: Result Delete(const std::string &path, const Headers &headers); Result Delete(const std::string &path, const char *body, size_t content_length, const std::string &content_type); + Result Delete(const std::string &path, const char *body, + size_t content_length, const std::string &content_type, + Progress progress); Result Delete(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type); + Result Delete(const std::string &path, const Headers &headers, + const char *body, size_t content_length, + const std::string &content_type, Progress progress); Result Delete(const std::string &path, const std::string &body, const std::string &content_type); + Result Delete(const std::string &path, const std::string &body, + const std::string &content_type, Progress progress); Result Delete(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type); + Result Delete(const std::string &path, const Headers &headers, + const std::string &body, const std::string &content_type, + Progress progress); Result Options(const std::string &path); Result Options(const std::string &path, const Headers &headers); @@ -1685,6 +1883,8 @@ public: #ifdef CPPHTTPLIB_OPENSSL_SUPPORT void enable_server_certificate_verification(bool enabled); + void enable_server_hostname_verification(bool enabled); + void set_server_certificate_verifier(std::function verifier); #endif void set_logger(Logger logger); @@ -1730,6 +1930,9 @@ public: SSL_CTX *ssl_context() const; + void update_certs(X509 *cert, EVP_PKEY *private_key, + X509_STORE *client_ca_cert_store = nullptr); + private: bool process_and_close_socket(socket_t sock) override; @@ -1810,68 +2013,58 @@ inline void duration_to_sec_and_usec(const T &duration, U callback) { callback(static_cast(sec), static_cast(usec)); } +inline bool is_numeric(const std::string &str) { + return !str.empty() && std::all_of(str.begin(), str.end(), ::isdigit); +} + inline uint64_t get_header_value_u64(const Headers &headers, - const std::string &key, size_t id, - uint64_t def) { + const std::string &key, uint64_t def, + size_t id, bool &is_invalid_value) { + is_invalid_value = false; auto rng = headers.equal_range(key); auto it = rng.first; std::advance(it, static_cast(id)); if (it != rng.second) { - return std::strtoull(it->second.data(), nullptr, 10); + if (is_numeric(it->second)) { + return std::strtoull(it->second.data(), nullptr, 10); + } else { + is_invalid_value = true; + } } return def; } +inline uint64_t get_header_value_u64(const Headers &headers, + const std::string &key, uint64_t def, + size_t id) { + bool dummy = false; + return get_header_value_u64(headers, key, def, id, dummy); +} + } // namespace detail inline uint64_t Request::get_header_value_u64(const std::string &key, - size_t id) const { - return detail::get_header_value_u64(headers, key, id, 0); + uint64_t def, size_t id) const { + return detail::get_header_value_u64(headers, key, def, id); } inline uint64_t Response::get_header_value_u64(const std::string &key, - size_t id) const { - return detail::get_header_value_u64(headers, key, id, 0); -} - -template -inline ssize_t Stream::write_format(const char *fmt, const Args &...args) { - const auto bufsiz = 2048; - std::array buf{}; - - auto sn = snprintf(buf.data(), buf.size() - 1, fmt, args...); - if (sn <= 0) { return sn; } - - auto n = static_cast(sn); - - if (n >= buf.size() - 1) { - std::vector glowable_buf(buf.size()); - - while (n >= glowable_buf.size() - 1) { - glowable_buf.resize(glowable_buf.size() * 2); - n = static_cast( - snprintf(&glowable_buf[0], glowable_buf.size() - 1, fmt, args...)); - } - return write(&glowable_buf[0], n); - } else { - return write(buf.data(), n); - } + uint64_t def, size_t id) const { + return detail::get_header_value_u64(headers, key, def, id); } inline void default_socket_options(socket_t sock) { - int yes = 1; + int opt = 1; #ifdef _WIN32 setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, - reinterpret_cast(&yes), sizeof(yes)); - setsockopt(sock, SOL_SOCKET, SO_EXCLUSIVEADDRUSE, - reinterpret_cast(&yes), sizeof(yes)); + reinterpret_cast(&opt), sizeof(opt)); #else #ifdef SO_REUSEPORT setsockopt(sock, SOL_SOCKET, SO_REUSEPORT, - reinterpret_cast(&yes), sizeof(yes)); + reinterpret_cast(&opt), sizeof(opt)); #else setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, - reinterpret_cast(&yes), sizeof(yes)); + reinterpret_cast(&opt), sizeof(opt)); #endif #endif } @@ -1997,6 +2190,8 @@ inline std::string to_string(const Error error) { case Error::SSLConnection: return "SSL connection failed"; case Error::SSLLoadingCerts: return "SSL certificate loading failed"; case Error::SSLServerVerification: return "SSL server verification failed"; + case Error::SSLServerHostnameVerification: + return "SSL server hostname verification failed"; case Error::UnsupportedMultipartBoundaryChars: return "Unsupported HTTP multipart boundary characters"; case Error::Compression: return "Compression failed"; @@ -2016,8 +2211,9 @@ inline std::ostream &operator<<(std::ostream &os, const Error &obj) { } inline uint64_t Result::get_request_header_value_u64(const std::string &key, + uint64_t def, size_t id) const { - return detail::get_header_value_u64(request_headers_, key, id, 0); + return detail::get_header_value_u64(request_headers_, key, def, id); } template @@ -2080,6 +2276,36 @@ make_basic_authentication_header(const std::string &username, namespace detail { +#if defined(_WIN32) +inline std::wstring u8string_to_wstring(const char *s) { + std::wstring ws; + auto len = static_cast(strlen(s)); + auto wlen = ::MultiByteToWideChar(CP_UTF8, 0, s, len, nullptr, 0); + if (wlen > 0) { + ws.resize(wlen); + wlen = ::MultiByteToWideChar( + CP_UTF8, 0, s, len, + const_cast(reinterpret_cast(ws.data())), wlen); + if (wlen != static_cast(ws.size())) { ws.clear(); } + } + return ws; +} +#endif + +struct FileStat { + FileStat(const std::string &path); + bool is_file() const; + bool is_dir() const; + +private: +#if defined(_WIN32) + struct _stat st_; +#else + struct stat st_; +#endif + int ret_ = -1; +}; + std::string encode_query_param(const std::string &value); std::string decode_url(const std::string &s, bool convert_plus_to_space); @@ -2088,6 +2314,16 @@ void read_file(const std::string &path, std::string &out); std::string trim_copy(const std::string &s); +void divide( + const char *data, std::size_t size, char d, + std::function + fn); + +void divide( + const std::string &str, char d, + std::function + fn); + void split(const char *b, const char *e, char d, std::function fn); @@ -2099,18 +2335,23 @@ bool process_client_socket(socket_t sock, time_t read_timeout_sec, time_t write_timeout_usec, std::function callback); -socket_t create_client_socket( - const std::string &host, const std::string &ip, int port, - int address_family, bool tcp_nodelay, SocketOptions socket_options, - time_t connection_timeout_sec, time_t connection_timeout_usec, - time_t read_timeout_sec, time_t read_timeout_usec, time_t write_timeout_sec, - time_t write_timeout_usec, const std::string &intf, Error &error); +socket_t create_client_socket(const std::string &host, const std::string &ip, + int port, int address_family, bool tcp_nodelay, + bool ipv6_v6only, SocketOptions socket_options, + time_t connection_timeout_sec, + time_t connection_timeout_usec, + time_t read_timeout_sec, time_t read_timeout_usec, + time_t write_timeout_sec, + time_t write_timeout_usec, + const std::string &intf, Error &error); const char *get_header_value(const Headers &headers, const std::string &key, - size_t id = 0, const char *def = nullptr); + const char *def, size_t id); std::string params_to_query_str(const Params ¶ms); +void parse_query_text(const char *data, std::size_t size, Params ¶ms); + void parse_query_text(const std::string &s, Params ¶ms); bool parse_multipart_boundary(const std::string &content_type, @@ -2270,15 +2511,70 @@ public: private: #if defined(_WIN32) - HANDLE hFile_; - HANDLE hMapping_; + HANDLE hFile_ = NULL; + HANDLE hMapping_ = NULL; #else - int fd_; + int fd_ = -1; #endif - size_t size_; - void *addr_; + size_t size_ = 0; + void *addr_ = nullptr; + bool is_open_empty_file = false; }; +// NOTE: https://www.rfc-editor.org/rfc/rfc9110#section-5 +namespace fields { + +inline bool is_token_char(char c) { + return std::isalnum(c) || c == '!' || c == '#' || c == '$' || c == '%' || + c == '&' || c == '\'' || c == '*' || c == '+' || c == '-' || + c == '.' || c == '^' || c == '_' || c == '`' || c == '|' || c == '~'; +} + +inline bool is_token(const std::string &s) { + if (s.empty()) { return false; } + for (auto c : s) { + if (!is_token_char(c)) { return false; } + } + return true; +} + +inline bool is_field_name(const std::string &s) { return is_token(s); } + +inline bool is_vchar(char c) { return c >= 33 && c <= 126; } + +inline bool is_obs_text(char c) { return 128 <= static_cast(c); } + +inline bool is_field_vchar(char c) { return is_vchar(c) || is_obs_text(c); } + +inline bool is_field_content(const std::string &s) { + if (s.empty()) { return false; } + + if (s.size() == 1) { + return is_field_vchar(s[0]); + } else if (s.size() == 2) { + return is_field_vchar(s[0]) && is_field_vchar(s[1]); + } else { + size_t i = 0; + + if (!is_field_vchar(s[i])) { return false; } + i++; + + while (i < s.size() - 1) { + auto c = s[i++]; + if (c == ' ' || c == '\t' || is_field_vchar(c)) { + } else { + return false; + } + } + + return is_field_vchar(s[i]); + } +} + +inline bool is_field_value(const std::string &s) { return is_field_content(s); } + +} // namespace fields + } // namespace detail // ---------------------------------------------------------------------------- @@ -2392,20 +2688,6 @@ inline std::string base64_encode(const std::string &in) { return out; } -inline bool is_file(const std::string &path) { -#ifdef _WIN32 - return _access_s(path.c_str(), 0) == 0; -#else - struct stat st; - return stat(path.c_str(), &st) >= 0 && S_ISREG(st.st_mode); -#endif -} - -inline bool is_dir(const std::string &path) { - struct stat st; - return stat(path.c_str(), &st) >= 0 && S_ISDIR(st.st_mode); -} - inline bool is_valid_path(const std::string &path) { size_t level = 0; size_t i = 0; @@ -2448,6 +2730,21 @@ inline bool is_valid_path(const std::string &path) { return true; } +inline FileStat::FileStat(const std::string &path) { +#if defined(_WIN32) + auto wpath = u8string_to_wstring(path.c_str()); + ret_ = _wstat(wpath.c_str(), &st_); +#else + ret_ = stat(path.c_str(), &st_); +#endif +} +inline bool FileStat::is_file() const { + return ret_ >= 0 && S_ISREG(st_.st_mode); +} +inline bool FileStat::is_dir() const { + return ret_ >= 0 && S_ISDIR(st_.st_mode); +} + inline std::string encode_query_param(const std::string &value) { std::ostringstream escaped; escaped.fill('0'); @@ -2579,6 +2876,27 @@ inline std::string trim_double_quotes_copy(const std::string &s) { return s; } +inline void +divide(const char *data, std::size_t size, char d, + std::function + fn) { + const auto it = std::find(data, data + size, d); + const auto found = static_cast(it != data + size); + const auto lhs_data = data; + const auto lhs_size = static_cast(it - data); + const auto rhs_data = it + found; + const auto rhs_size = size - lhs_size - found; + + fn(lhs_data, lhs_size, rhs_data, rhs_size); +} + +inline void +divide(const std::string &str, char d, + std::function + fn) { + divide(str.data(), str.size(), d, std::move(fn)); +} + inline void split(const char *b, const char *e, char d, std::function fn) { return split(b, e, d, (std::numeric_limits::max)(), std::move(fn)); @@ -2636,6 +2954,10 @@ inline bool stream_line_reader::getline() { fixed_buffer_used_size_ = 0; glowable_buffer_.clear(); +#ifndef CPPHTTPLIB_ALLOW_LF_AS_LINE_TERMINATOR + char prev_byte = 0; +#endif + for (size_t i = 0;; i++) { char byte; auto n = strm_.read(&byte, 1); @@ -2652,7 +2974,12 @@ inline bool stream_line_reader::getline() { append(byte); +#ifdef CPPHTTPLIB_ALLOW_LF_AS_LINE_TERMINATOR if (byte == '\n') { break; } +#else + if (prev_byte == '\r' && byte == '\n') { break; } + prev_byte = byte; +#endif } return true; @@ -2671,16 +2998,7 @@ inline void stream_line_reader::append(char c) { } } -inline mmap::mmap(const char *path) -#if defined(_WIN32) - : hFile_(NULL), hMapping_(NULL) -#else - : fd_(-1) -#endif - , - size_(0), addr_(nullptr) { - open(path); -} +inline mmap::mmap(const char *path) { open(path); } inline mmap::~mmap() { close(); } @@ -2688,29 +3006,60 @@ inline bool mmap::open(const char *path) { close(); #if defined(_WIN32) - std::wstring wpath; - for (size_t i = 0; i < strlen(path); i++) { - wpath += path[i]; - } + auto wpath = u8string_to_wstring(path); + if (wpath.empty()) { return false; } +#if _WIN32_WINNT >= _WIN32_WINNT_WIN8 hFile_ = ::CreateFile2(wpath.c_str(), GENERIC_READ, FILE_SHARE_READ, OPEN_EXISTING, NULL); +#else + hFile_ = ::CreateFileW(wpath.c_str(), GENERIC_READ, FILE_SHARE_READ, NULL, + OPEN_EXISTING, FILE_ATTRIBUTE_NORMAL, NULL); +#endif if (hFile_ == INVALID_HANDLE_VALUE) { return false; } LARGE_INTEGER size{}; if (!::GetFileSizeEx(hFile_, &size)) { return false; } + // If the following line doesn't compile due to QuadPart, update Windows SDK. + // See: + // https://github.com/yhirose/cpp-httplib/issues/1903#issuecomment-2316520721 + if (static_cast(size.QuadPart) > + (std::numeric_limits::max)()) { + // `size_t` might be 32-bits, on 32-bits Windows. + return false; + } size_ = static_cast(size.QuadPart); +#if _WIN32_WINNT >= _WIN32_WINNT_WIN8 hMapping_ = ::CreateFileMappingFromApp(hFile_, NULL, PAGE_READONLY, size_, NULL); +#else + hMapping_ = ::CreateFileMappingW(hFile_, NULL, PAGE_READONLY, 0, 0, NULL); +#endif + + // Special treatment for an empty file... + if (hMapping_ == NULL && size_ == 0) { + close(); + is_open_empty_file = true; + return true; + } if (hMapping_ == NULL) { close(); return false; } +#if _WIN32_WINNT >= _WIN32_WINNT_WIN8 addr_ = ::MapViewOfFileFromApp(hMapping_, FILE_MAP_READ, 0, 0); +#else + addr_ = ::MapViewOfFile(hMapping_, FILE_MAP_READ, 0, 0, 0); +#endif + + if (addr_ == nullptr) { + close(); + return false; + } #else fd_ = ::open(path, O_RDONLY); if (fd_ == -1) { return false; } @@ -2723,22 +3072,26 @@ inline bool mmap::open(const char *path) { size_ = static_cast(sb.st_size); addr_ = ::mmap(NULL, size_, PROT_READ, MAP_PRIVATE, fd_, 0); -#endif - if (addr_ == nullptr) { + // Special treatment for an empty file... + if (addr_ == MAP_FAILED && size_ == 0) { close(); + is_open_empty_file = true; return false; } +#endif return true; } -inline bool mmap::is_open() const { return addr_ != nullptr; } +inline bool mmap::is_open() const { + return is_open_empty_file ? true : addr_ != nullptr; +} inline size_t mmap::size() const { return size_; } inline const char *mmap::data() const { - return static_cast(addr_); + return is_open_empty_file ? "" : static_cast(addr_); } inline void mmap::close() { @@ -2757,6 +3110,8 @@ inline void mmap::close() { ::CloseHandle(hFile_); hFile_ = INVALID_HANDLE_VALUE; } + + is_open_empty_file = false; #else if (addr_ != nullptr) { munmap(addr_, size_); @@ -2782,7 +3137,10 @@ template inline ssize_t handle_EINTR(T fn) { ssize_t res = 0; while (true) { res = fn(); - if (res < 0 && errno == EINTR) { continue; } + if (res < 0 && errno == EINTR) { + std::this_thread::sleep_for(std::chrono::microseconds{1}); + continue; + } break; } return res; @@ -2991,23 +3349,37 @@ private: }; #endif -inline bool keep_alive(socket_t sock, time_t keep_alive_timeout_sec) { +inline bool keep_alive(const std::atomic &svr_sock, socket_t sock, + time_t keep_alive_timeout_sec) { using namespace std::chrono; - auto start = steady_clock::now(); + + const auto interval_usec = + CPPHTTPLIB_KEEPALIVE_TIMEOUT_CHECK_INTERVAL_USECOND; + + // Avoid expensive `steady_clock::now()` call for the first time + if (select_read(sock, 0, interval_usec) > 0) { return true; } + + const auto start = steady_clock::now() - microseconds{interval_usec}; + const auto timeout = seconds{keep_alive_timeout_sec}; + while (true) { - auto val = select_read(sock, 0, 10000); + if (svr_sock == INVALID_SOCKET) { + break; // Server socket is closed + } + + auto val = select_read(sock, 0, interval_usec); if (val < 0) { - return false; + break; // Ssocket error } else if (val == 0) { - auto current = steady_clock::now(); - auto duration = duration_cast(current - start); - auto timeout = keep_alive_timeout_sec * 1000; - if (duration.count() > timeout) { return false; } - std::this_thread::sleep_for(std::chrono::milliseconds(1)); + if (steady_clock::now() - start > timeout) { + break; // Timeout + } } else { - return true; + return true; // Ready for read } } + + return false; } template @@ -3018,8 +3390,7 @@ process_server_socket_core(const std::atomic &svr_sock, socket_t sock, assert(keep_alive_max_count > 0); auto ret = false; auto count = keep_alive_max_count; - while (svr_sock != INVALID_SOCKET && count > 0 && - keep_alive(sock, keep_alive_timeout_sec)) { + while (count > 0 && keep_alive(svr_sock, sock, keep_alive_timeout_sec)) { auto close_connection = count == 1; auto connection_closed = false; ret = callback(close_connection, connection_closed); @@ -3063,10 +3434,29 @@ inline int shutdown_socket(socket_t sock) { #endif } +inline std::string escape_abstract_namespace_unix_domain(const std::string &s) { + if (s.size() > 1 && s[0] == '\0') { + auto ret = s; + ret[0] = '@'; + return ret; + } + return s; +} + +inline std::string +unescape_abstract_namespace_unix_domain(const std::string &s) { + if (s.size() > 1 && s[0] == '@') { + auto ret = s; + ret[0] = '\0'; + return ret; + } + return s; +} + template socket_t create_socket(const std::string &host, const std::string &ip, int port, int address_family, int socket_flags, bool tcp_nodelay, - SocketOptions socket_options, + bool ipv6_v6only, SocketOptions socket_options, BindOrConnect bind_or_connect) { // Get address info const char *node = nullptr; @@ -3075,7 +3465,7 @@ socket_t create_socket(const std::string &host, const std::string &ip, int port, memset(&hints, 0, sizeof(struct addrinfo)); hints.ai_socktype = SOCK_STREAM; - hints.ai_protocol = 0; + hints.ai_protocol = IPPROTO_IP; if (!ip.empty()) { node = ip.c_str(); @@ -3093,20 +3483,32 @@ socket_t create_socket(const std::string &host, const std::string &ip, int port, const auto addrlen = host.length(); if (addrlen > sizeof(sockaddr_un::sun_path)) { return INVALID_SOCKET; } +#ifdef SOCK_CLOEXEC + auto sock = socket(hints.ai_family, hints.ai_socktype | SOCK_CLOEXEC, + hints.ai_protocol); +#else auto sock = socket(hints.ai_family, hints.ai_socktype, hints.ai_protocol); +#endif + if (sock != INVALID_SOCKET) { sockaddr_un addr{}; addr.sun_family = AF_UNIX; - std::copy(host.begin(), host.end(), addr.sun_path); + + auto unescaped_host = unescape_abstract_namespace_unix_domain(host); + std::copy(unescaped_host.begin(), unescaped_host.end(), addr.sun_path); hints.ai_addr = reinterpret_cast(&addr); hints.ai_addrlen = static_cast( sizeof(addr) - sizeof(addr.sun_path) + addrlen); +#ifndef SOCK_CLOEXEC fcntl(sock, F_SETFD, FD_CLOEXEC); +#endif + if (socket_options) { socket_options(sock); } - if (!bind_or_connect(sock, hints)) { + bool dummy; + if (!bind_or_connect(sock, hints, dummy)) { close_socket(sock); sock = INVALID_SOCKET; } @@ -3123,6 +3525,7 @@ socket_t create_socket(const std::string &host, const std::string &ip, int port, #endif return INVALID_SOCKET; } + auto se = detail::scope_exit([&] { freeaddrinfo(result); }); for (auto rp = result; rp; rp = rp->ai_next) { // Create a socket @@ -3148,11 +3551,18 @@ socket_t create_socket(const std::string &host, const std::string &ip, int port, sock = socket(rp->ai_family, rp->ai_socktype, rp->ai_protocol); } #else + +#ifdef SOCK_CLOEXEC + auto sock = + socket(rp->ai_family, rp->ai_socktype | SOCK_CLOEXEC, rp->ai_protocol); +#else auto sock = socket(rp->ai_family, rp->ai_socktype, rp->ai_protocol); +#endif + #endif if (sock == INVALID_SOCKET) { continue; } -#ifndef _WIN32 +#if !defined _WIN32 && !defined SOCK_CLOEXEC if (fcntl(sock, F_SETFD, FD_CLOEXEC) == -1) { close_socket(sock); continue; @@ -3160,39 +3570,38 @@ socket_t create_socket(const std::string &host, const std::string &ip, int port, #endif if (tcp_nodelay) { - auto yes = 1; + auto opt = 1; #ifdef _WIN32 setsockopt(sock, IPPROTO_TCP, TCP_NODELAY, - reinterpret_cast(&yes), sizeof(yes)); + reinterpret_cast(&opt), sizeof(opt)); #else setsockopt(sock, IPPROTO_TCP, TCP_NODELAY, - reinterpret_cast(&yes), sizeof(yes)); + reinterpret_cast(&opt), sizeof(opt)); +#endif + } + + if (rp->ai_family == AF_INET6) { + auto opt = ipv6_v6only ? 1 : 0; +#ifdef _WIN32 + setsockopt(sock, IPPROTO_IPV6, IPV6_V6ONLY, + reinterpret_cast(&opt), sizeof(opt)); +#else + setsockopt(sock, IPPROTO_IPV6, IPV6_V6ONLY, + reinterpret_cast(&opt), sizeof(opt)); #endif } if (socket_options) { socket_options(sock); } - if (rp->ai_family == AF_INET6) { - auto no = 0; -#ifdef _WIN32 - setsockopt(sock, IPPROTO_IPV6, IPV6_V6ONLY, - reinterpret_cast(&no), sizeof(no)); -#else - setsockopt(sock, IPPROTO_IPV6, IPV6_V6ONLY, - reinterpret_cast(&no), sizeof(no)); -#endif - } - // bind or connect - if (bind_or_connect(sock, *rp)) { - freeaddrinfo(result); - return sock; - } + auto quit = false; + if (bind_or_connect(sock, *rp, quit)) { return sock; } close_socket(sock); + + if (quit) { break; } } - freeaddrinfo(result); return INVALID_SOCKET; } @@ -3225,6 +3634,7 @@ inline bool bind_ip_address(socket_t sock, const std::string &host) { hints.ai_protocol = 0; if (getaddrinfo(host.c_str(), "0", &hints, &result)) { return false; } + auto se = detail::scope_exit([&] { freeaddrinfo(result); }); auto ret = false; for (auto rp = result; rp; rp = rp->ai_next) { @@ -3235,7 +3645,6 @@ inline bool bind_ip_address(socket_t sock, const std::string &host) { } } - freeaddrinfo(result); return ret; } @@ -3247,6 +3656,8 @@ inline bool bind_ip_address(socket_t sock, const std::string &host) { inline std::string if2ip(int address_family, const std::string &ifn) { struct ifaddrs *ifap; getifaddrs(&ifap); + auto se = detail::scope_exit([&] { freeifaddrs(ifap); }); + std::string addr_candidate; for (auto ifa = ifap; ifa; ifa = ifa->ifa_next) { if (ifa->ifa_addr && ifn == ifa->ifa_name && @@ -3256,7 +3667,6 @@ inline std::string if2ip(int address_family, const std::string &ifn) { auto sa = reinterpret_cast(ifa->ifa_addr); char buf[INET_ADDRSTRLEN]; if (inet_ntop(AF_INET, &sa->sin_addr, buf, INET_ADDRSTRLEN)) { - freeifaddrs(ifap); return std::string(buf, INET_ADDRSTRLEN); } } else if (ifa->ifa_addr->sa_family == AF_INET6) { @@ -3269,7 +3679,6 @@ inline std::string if2ip(int address_family, const std::string &ifn) { if (s6_addr_head == 0xfc || s6_addr_head == 0xfd) { addr_candidate = std::string(buf, INET6_ADDRSTRLEN); } else { - freeifaddrs(ifap); return std::string(buf, INET6_ADDRSTRLEN); } } @@ -3277,20 +3686,21 @@ inline std::string if2ip(int address_family, const std::string &ifn) { } } } - freeifaddrs(ifap); return addr_candidate; } #endif inline socket_t create_client_socket( const std::string &host, const std::string &ip, int port, - int address_family, bool tcp_nodelay, SocketOptions socket_options, - time_t connection_timeout_sec, time_t connection_timeout_usec, - time_t read_timeout_sec, time_t read_timeout_usec, time_t write_timeout_sec, + int address_family, bool tcp_nodelay, bool ipv6_v6only, + SocketOptions socket_options, time_t connection_timeout_sec, + time_t connection_timeout_usec, time_t read_timeout_sec, + time_t read_timeout_usec, time_t write_timeout_sec, time_t write_timeout_usec, const std::string &intf, Error &error) { auto sock = create_socket( - host, ip, port, address_family, 0, tcp_nodelay, std::move(socket_options), - [&](socket_t sock2, struct addrinfo &ai) -> bool { + host, ip, port, address_family, 0, tcp_nodelay, ipv6_v6only, + std::move(socket_options), + [&](socket_t sock2, struct addrinfo &ai, bool &quit) -> bool { if (!intf.empty()) { #ifdef USE_IF2IP auto ip_from_if = if2ip(address_family, intf); @@ -3314,7 +3724,10 @@ inline socket_t create_client_socket( } error = wait_until_socket_is_ready(sock2, connection_timeout_sec, connection_timeout_usec); - if (error != Error::Success) { return false; } + if (error != Error::Success) { + if (error == Error::ConnectionTimeout) { quit = true; } + return false; + } } set_nonblocking(sock2, false); @@ -3439,7 +3852,7 @@ inline unsigned int str2tag(const std::string &s) { namespace udl { -inline constexpr unsigned int operator"" _t(const char *s, size_t l) { +inline constexpr unsigned int operator""_t(const char *s, size_t l) { return str2tag_core(s, l, 0); } @@ -3524,8 +3937,9 @@ inline bool can_compress_content_type(const std::string &content_type) { case "application/protobuf"_t: case "application/xhtml+xml"_t: return true; - default: - return !content_type.rfind("text/", 0) && tag != "text/event-stream"_t; + case "text/event-stream"_t: return false; + + default: return !content_type.rfind("text/", 0); } } @@ -3762,8 +4176,8 @@ inline bool has_header(const Headers &headers, const std::string &key) { } inline const char *get_header_value(const Headers &headers, - const std::string &key, size_t id, - const char *def) { + const std::string &key, const char *def, + size_t id) { auto rng = headers.equal_range(key); auto it = rng.first; std::advance(it, static_cast(id)); @@ -3771,14 +4185,6 @@ inline const char *get_header_value(const Headers &headers, return def; } -inline bool compare_case_ignore(const std::string &a, const std::string &b) { - if (a.size() != b.size()) { return false; } - for (size_t i = 0; i < b.size(); i++) { - if (::tolower(a[i]) != ::tolower(b[i])) { return false; } - } - return true; -} - template inline bool parse_header(const char *beg, const char *end, T fn) { // Skip trailing spaces and tabs. @@ -3801,15 +4207,27 @@ inline bool parse_header(const char *beg, const char *end, T fn) { p++; } - if (p < end) { + if (p <= end) { auto key_len = key_end - beg; if (!key_len) { return false; } auto key = std::string(beg, key_end); - auto val = compare_case_ignore(key, "Location") + auto val = case_ignore::equal(key, "Location") ? std::string(p, end) : decode_url(std::string(p, end), false); - fn(std::move(key), std::move(val)); + + // NOTE: From RFC 9110: + // Field values containing CR, LF, or NUL characters are + // invalid and dangerous, due to the varying ways that + // implementations might parse and interpret those + // characters; a recipient of CR, LF, or NUL within a field + // value MUST either reject the message or replace each of + // those characters with SP before further processing or + // forwarding of that message. + static const std::string CR_LF_NUL("\r\n\0", 3); + if (val.find_first_of(CR_LF_NUL) != std::string::npos) { return false; } + + fn(key, val); return true; } @@ -3829,27 +4247,27 @@ inline bool read_headers(Stream &strm, Headers &headers) { if (line_reader.end_with_crlf()) { // Blank line indicates end of headers. if (line_reader.size() == 2) { break; } -#ifdef CPPHTTPLIB_ALLOW_LF_AS_LINE_TERMINATOR } else { +#ifdef CPPHTTPLIB_ALLOW_LF_AS_LINE_TERMINATOR // Blank line indicates end of headers. if (line_reader.size() == 1) { break; } line_terminator_len = 1; - } #else - } else { continue; // Skip invalid line. - } #endif + } if (line_reader.size() > CPPHTTPLIB_HEADER_MAX_LENGTH) { return false; } // Exclude line terminator auto end = line_reader.ptr() + line_reader.size() - line_terminator_len; - parse_header(line_reader.ptr(), end, - [&](std::string &&key, std::string &&val) { - headers.emplace(std::move(key), std::move(val)); - }); + if (!parse_header(line_reader.ptr(), end, + [&](const std::string &key, std::string &val) { + headers.emplace(key, val); + })) { + return false; + } } return true; @@ -3937,8 +4355,19 @@ inline bool read_content_chunked(Stream &strm, T &x, assert(chunk_len == 0); - // Trailer - if (!line_reader.getline()) { return false; } + // NOTE: In RFC 9112, '7.1 Chunked Transfer Coding' mentiones "The chunked + // transfer coding is complete when a chunk with a chunk-size of zero is + // received, possibly followed by a trailer section, and finally terminated by + // an empty line". https://www.rfc-editor.org/rfc/rfc9112.html#section-7.1 + // + // In '7.1.3. Decoding Chunked', however, the pseudo-code in the section + // does't care for the existence of the final CRLF. In other words, it seems + // to be ok whether the final CRLF exists or not in the chunked data. + // https://www.rfc-editor.org/rfc/rfc9112.html#section-7.1.3 + // + // According to the reference code in RFC 9112, cpp-htpplib now allows + // chuncked transfer coding data without the final CRLF. + if (!line_reader.getline()) { return true; } while (strcmp(line_reader.ptr(), "\r\n") != 0) { if (line_reader.size() > CPPHTTPLIB_HEADER_MAX_LENGTH) { return false; } @@ -3948,8 +4377,8 @@ inline bool read_content_chunked(Stream &strm, T &x, auto end = line_reader.ptr() + line_reader.size() - line_terminator_len; parse_header(line_reader.ptr(), end, - [&](std::string &&key, std::string &&val) { - x.headers.emplace(std::move(key), std::move(val)); + [&](const std::string &key, const std::string &val) { + x.headers.emplace(key, val); }); if (!line_reader.getline()) { return false; } @@ -3959,8 +4388,8 @@ inline bool read_content_chunked(Stream &strm, T &x, } inline bool is_chunked_transfer_encoding(const Headers &headers) { - return compare_case_ignore( - get_header_value(headers, "Transfer-Encoding", 0, ""), "chunked"); + return case_ignore::equal( + get_header_value(headers, "Transfer-Encoding", "", 0), "chunked"); } template @@ -4026,8 +4455,14 @@ bool read_content(Stream &strm, T &x, size_t payload_max_length, int &status, } else if (!has_header(x.headers, "Content-Length")) { ret = read_content_without_length(strm, out); } else { - auto len = get_header_value_u64(x.headers, "Content-Length", 0, 0); - if (len > payload_max_length) { + auto is_invalid_value = false; + auto len = get_header_value_u64(x.headers, "Content-Length", + std::numeric_limits::max(), + 0, is_invalid_value); + + if (is_invalid_value) { + ret = false; + } else if (len > payload_max_length) { exceed_payload_max_length = true; skip_content_with_length(strm, len); ret = false; @@ -4042,13 +4477,36 @@ bool read_content(Stream &strm, T &x, size_t payload_max_length, int &status, } return ret; }); -} // namespace detail +} + +inline ssize_t write_request_line(Stream &strm, const std::string &method, + const std::string &path) { + std::string s = method; + s += " "; + s += path; + s += " HTTP/1.1\r\n"; + return strm.write(s.data(), s.size()); +} + +inline ssize_t write_response_line(Stream &strm, int status) { + std::string s = "HTTP/1.1 "; + s += std::to_string(status); + s += " "; + s += httplib::status_message(status); + s += "\r\n"; + return strm.write(s.data(), s.size()); +} inline ssize_t write_headers(Stream &strm, const Headers &headers) { ssize_t write_len = 0; for (const auto &x : headers) { - auto len = - strm.write_format("%s: %s\r\n", x.first.c_str(), x.second.c_str()); + std::string s; + s = x.first; + s += ": "; + s += x.second; + s += "\r\n"; + + auto len = strm.write(s.data(), s.size()); if (len < 0) { return len; } write_len += len; } @@ -4302,22 +4760,22 @@ inline std::string params_to_query_str(const Params ¶ms) { return query; } -inline void parse_query_text(const std::string &s, Params ¶ms) { +inline void parse_query_text(const char *data, std::size_t size, + Params ¶ms) { std::set cache; - split(s.data(), s.data() + s.size(), '&', [&](const char *b, const char *e) { + split(data, data + size, '&', [&](const char *b, const char *e) { std::string kv(b, e); if (cache.find(kv) != cache.end()) { return; } - cache.insert(kv); + cache.insert(std::move(kv)); std::string key; std::string val; - split(b, e, '=', [&](const char *b2, const char *e2) { - if (key.empty()) { - key.assign(b2, e2); - } else { - val.assign(b2, e2); - } - }); + divide(b, static_cast(e - b), '=', + [&](const char *lhs_data, std::size_t lhs_size, const char *rhs_data, + std::size_t rhs_size) { + key.assign(lhs_data, lhs_size); + val.assign(rhs_data, rhs_size); + }); if (!key.empty()) { params.emplace(decode_url(key, true), decode_url(val, true)); @@ -4325,6 +4783,10 @@ inline void parse_query_text(const std::string &s, Params ¶ms) { }); } +inline void parse_query_text(const std::string &s, Params ¶ms) { + parse_query_text(s.data(), s.size(), params); +} + inline bool parse_multipart_boundary(const std::string &content_type, std::string &boundary) { auto boundary_keyword = "boundary="; @@ -4365,35 +4827,44 @@ inline bool parse_range_header(const std::string &s, Ranges &ranges) { #else inline bool parse_range_header(const std::string &s, Ranges &ranges) try { #endif - static auto re_first_range = std::regex(R"(bytes=(\d*-\d*(?:,\s*\d*-\d*)*))"); - std::smatch m; - if (std::regex_match(s, m, re_first_range)) { - auto pos = static_cast(m.position(1)); - auto len = static_cast(m.length(1)); + auto is_valid = [](const std::string &str) { + return std::all_of(str.cbegin(), str.cend(), + [](unsigned char c) { return std::isdigit(c); }); + }; + + if (s.size() > 7 && s.compare(0, 6, "bytes=") == 0) { + const auto pos = static_cast(6); + const auto len = static_cast(s.size() - 6); auto all_valid_ranges = true; split(&s[pos], &s[pos + len], ',', [&](const char *b, const char *e) { if (!all_valid_ranges) { return; } - static auto re_another_range = std::regex(R"(\s*(\d*)-(\d*))"); - std::cmatch cm; - if (std::regex_match(b, e, cm, re_another_range)) { - ssize_t first = -1; - if (!cm.str(1).empty()) { - first = static_cast(std::stoll(cm.str(1))); - } - ssize_t last = -1; - if (!cm.str(2).empty()) { - last = static_cast(std::stoll(cm.str(2))); - } - - if (first != -1 && last != -1 && first > last) { - all_valid_ranges = false; - return; - } - ranges.emplace_back(std::make_pair(first, last)); + const auto it = std::find(b, e, '-'); + if (it == e) { + all_valid_ranges = false; + return; } + + const auto lhs = std::string(b, it); + const auto rhs = std::string(it + 1, e); + if (!is_valid(lhs) || !is_valid(rhs)) { + all_valid_ranges = false; + return; + } + + const auto first = + static_cast(lhs.empty() ? -1 : std::stoll(lhs)); + const auto last = + static_cast(rhs.empty() ? -1 : std::stoll(rhs)); + if ((first == -1 && last == -1) || + (first != -1 && last != -1 && first > last)) { + all_valid_ranges = false; + return; + } + + ranges.emplace_back(first, last); }); - return all_valid_ranges; + return all_valid_ranges && !ranges.empty(); } return false; #ifdef CPPHTTPLIB_NO_EXCEPTIONS @@ -4452,7 +4923,7 @@ public: const auto header = buf_head(pos); if (!parse_header(header.data(), header.data() + header.size(), - [&](std::string &&, std::string &&) {})) { + [&](const std::string &, const std::string &) {})) { is_valid_ = false; return false; } @@ -4562,7 +5033,9 @@ private: const std::string &b) const { if (a.size() < b.size()) { return false; } for (size_t i = 0; i < b.size(); i++) { - if (::tolower(a[i]) != ::tolower(b[i])) { return false; } + if (case_ignore::to_lower(a[i]) != case_ignore::to_lower(b[i])) { + return false; + } } return true; } @@ -4645,16 +5118,6 @@ private: size_t buf_epos_ = 0; }; -inline std::string to_lower(const char *beg, const char *end) { - std::string out; - auto it = beg; - while (it != end) { - out += static_cast(::tolower(*it)); - it++; - } - return out; -} - inline std::string random_string(size_t length) { static const char data[] = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz"; @@ -4768,7 +5231,18 @@ inline bool range_error(Request &req, Response &res) { last_pos = contant_len - 1; } - if (last_pos == -1) { last_pos = contant_len - 1; } + // NOTE: RFC-9110 '14.1.2. Byte Ranges': + // A client can limit the number of bytes requested without knowing the + // size of the selected representation. If the last-pos value is absent, + // or if the value is greater than or equal to the current length of the + // representation data, the byte range is interpreted as the remainder of + // the representation (i.e., the server replaces the value of last-pos + // with a value that is one less than the current length of the selected + // representation). + // https://www.rfc-editor.org/rfc/rfc9110.html#section-14.1.2-6 + if (last_pos == -1 || last_pos >= contant_len) { + last_pos = contant_len - 1; + } // Range must be within content length if (!(0 <= first_pos && first_pos <= last_pos && @@ -4795,12 +5269,11 @@ inline bool range_error(Request &req, Response &res) { inline std::pair get_range_offset_and_length(Range r, size_t content_length) { - (void)(content_length); // patch to get rid of "unused parameter" on release build assert(r.first != -1 && r.second != -1); assert(0 <= r.first && r.first < static_cast(content_length)); assert(r.first <= r.second && r.second < static_cast(content_length)); - + (void)(content_length); return std::make_pair(r.first, static_cast(r.second - r.first) + 1); } @@ -5230,6 +5703,7 @@ inline void hosted_at(const std::string &hostname, #endif return; } + auto se = detail::scope_exit([&] { freeaddrinfo(result); }); for (auto rp = result; rp; rp = rp->ai_next) { const auto &addr = @@ -5241,8 +5715,6 @@ inline void hosted_at(const std::string &hostname, addrs.push_back(ip); } } - - freeaddrinfo(result); } inline std::string append_query_params(const std::string &path, @@ -5291,8 +5763,8 @@ inline bool Request::has_header(const std::string &key) const { } inline std::string Request::get_header_value(const std::string &key, - size_t id) const { - return detail::get_header_value(headers, key, id, ""); + const char *def, size_t id) const { + return detail::get_header_value(headers, key, def, id); } inline size_t Request::get_header_value_count(const std::string &key) const { @@ -5302,7 +5774,8 @@ inline size_t Request::get_header_value_count(const std::string &key) const { inline void Request::set_header(const std::string &key, const std::string &val) { - if (!detail::has_crlf(key) && !detail::has_crlf(val)) { + if (detail::fields::is_field_name(key) && + detail::fields::is_field_value(val)) { headers.emplace(key, val); } } @@ -5356,8 +5829,9 @@ inline bool Response::has_header(const std::string &key) const { } inline std::string Response::get_header_value(const std::string &key, + const char *def, size_t id) const { - return detail::get_header_value(headers, key, id, ""); + return detail::get_header_value(headers, key, def, id); } inline size_t Response::get_header_value_count(const std::string &key) const { @@ -5367,13 +5841,14 @@ inline size_t Response::get_header_value_count(const std::string &key) const { inline void Response::set_header(const std::string &key, const std::string &val) { - if (!detail::has_crlf(key) && !detail::has_crlf(val)) { + if (detail::fields::is_field_name(key) && + detail::fields::is_field_value(val)) { headers.emplace(key, val); } } inline void Response::set_redirect(const std::string &url, int stat) { - if (!detail::has_crlf(url)) { + if (detail::fields::is_field_value(url)) { set_header("Location", url); if (300 <= stat && stat < 400) { this->status = stat; @@ -5436,14 +5911,25 @@ inline void Response::set_chunked_content_provider( is_chunked_content_provider_ = true; } +inline void Response::set_file_content(const std::string &path, + const std::string &content_type) { + file_content_path_ = path; + file_content_content_type_ = content_type; +} + +inline void Response::set_file_content(const std::string &path) { + file_content_path_ = path; +} + // Result implementation inline bool Result::has_request_header(const std::string &key) const { return request_headers_.find(key) != request_headers_.end(); } inline std::string Result::get_request_header_value(const std::string &key, + const char *def, size_t id) const { - return detail::get_header_value(request_headers_, key, id, ""); + return detail::get_header_value(request_headers_, key, def, id); } inline size_t @@ -5584,6 +6070,8 @@ inline socket_t BufferStream::socket() const { return 0; } inline const std::string &BufferStream::get_buffer() const { return buffer; } inline PathParamsMatcher::PathParamsMatcher(const std::string &pattern) { + static constexpr char marker[] = "/:"; + // One past the last ending position of a path param substring std::size_t last_param_end = 0; @@ -5596,13 +6084,14 @@ inline PathParamsMatcher::PathParamsMatcher(const std::string &pattern) { #endif while (true) { - const auto marker_pos = pattern.find(marker, last_param_end); + const auto marker_pos = pattern.find( + marker, last_param_end == 0 ? last_param_end : last_param_end - 1); if (marker_pos == std::string::npos) { break; } static_fragments_.push_back( - pattern.substr(last_param_end, marker_pos - last_param_end)); + pattern.substr(last_param_end, marker_pos - last_param_end + 1)); - const auto param_name_start = marker_pos + 1; + const auto param_name_start = marker_pos + 2; auto sep_pos = pattern.find(separator, param_name_start); if (sep_pos == std::string::npos) { sep_pos = pattern.length(); } @@ -5664,7 +6153,7 @@ inline bool PathParamsMatcher::match(Request &request) const { request.path_params.emplace( param_name, request.path.substr(starting_pos, sep_pos - starting_pos)); - // Mark everythin up to '/' as matched + // Mark everything up to '/' as matched starting_pos = sep_pos + 1; } // Returns false if the path is longer than the pattern @@ -5763,7 +6252,8 @@ inline bool Server::set_base_dir(const std::string &dir, inline bool Server::set_mount_point(const std::string &mount_point, const std::string &dir, Headers headers) { - if (detail::is_dir(dir)) { + detail::FileStat stat(dir); + if (stat.is_dir()) { std::string mnt = !mount_point.empty() ? mount_point : "/"; if (!mnt.empty() && mnt[0] == '/') { base_dirs_.push_back({mnt, dir, std::move(headers)}); @@ -5800,12 +6290,14 @@ inline Server &Server::set_file_request_handler(Handler handler) { return *this; } -inline Server &Server::set_error_handler(HandlerWithResponse handler) { +inline Server &Server::set_error_handler_core(HandlerWithResponse handler, + std::true_type) { error_handler_ = std::move(handler); return *this; } -inline Server &Server::set_error_handler(Handler handler) { +inline Server &Server::set_error_handler_core(Handler handler, + std::false_type) { error_handler_ = [handler](const Request &req, Response &res) { handler(req, res); return HandlerResponse::Handled; @@ -5849,6 +6341,11 @@ inline Server &Server::set_tcp_nodelay(bool on) { return *this; } +inline Server &Server::set_ipv6_v6only(bool on) { + ipv6_v6only_ = on; + return *this; +} + inline Server &Server::set_socket_options(SocketOptions socket_options) { socket_options_ = std::move(socket_options); return *this; @@ -5900,27 +6397,27 @@ inline Server &Server::set_payload_max_length(size_t length) { inline bool Server::bind_to_port(const std::string &host, int port, int socket_flags) { - return bind_internal(host, port, socket_flags) >= 0; + auto ret = bind_internal(host, port, socket_flags); + if (ret == -1) { is_decommisioned = true; } + return ret >= 0; } inline int Server::bind_to_any_port(const std::string &host, int socket_flags) { - return bind_internal(host, 0, socket_flags); + auto ret = bind_internal(host, 0, socket_flags); + if (ret == -1) { is_decommisioned = true; } + return ret; } -inline bool Server::listen_after_bind() { - auto se = detail::scope_exit([&]() { done_ = true; }); - return listen_internal(); -} +inline bool Server::listen_after_bind() { return listen_internal(); } inline bool Server::listen(const std::string &host, int port, int socket_flags) { - auto se = detail::scope_exit([&]() { done_ = true; }); return bind_to_port(host, port, socket_flags) && listen_internal(); } inline bool Server::is_running() const { return is_running_; } inline void Server::wait_until_ready() const { - while (!is_running() && !done_) { + while (!is_running_ && !is_decommisioned) { std::this_thread::sleep_for(std::chrono::milliseconds{1}); } } @@ -5932,8 +6429,11 @@ inline void Server::stop() { detail::shutdown_socket(sock); detail::close_socket(sock); } + is_decommisioned = false; } +inline void Server::decommission() { is_decommisioned = true; } + inline bool Server::parse_request_line(const char *s, Request &req) const { auto len = strlen(s); if (len < 2 || s[len - 2] != '\r' || s[len - 1] != '\n') { return false; } @@ -5972,26 +6472,13 @@ inline bool Server::parse_request_line(const char *s, Request &req) const { } } - size_t count = 0; - - detail::split(req.target.data(), req.target.data() + req.target.size(), '?', - 2, [&](const char *b, const char *e) { - switch (count) { - case 0: - req.path = detail::decode_url(std::string(b, e), false); - break; - case 1: { - if (e - b > 0) { - detail::parse_query_text(std::string(b, e), req.params); - } - break; - } - default: break; - } - count++; - }); - - if (count > 2) { return false; } + detail::divide(req.target, '?', + [&](const char *lhs_data, std::size_t lhs_size, + const char *rhs_data, std::size_t rhs_size) { + req.path = detail::decode_url( + std::string(lhs_data, lhs_size), false); + detail::parse_query_text(rhs_data, rhs_size, req.params); + }); } return true; @@ -6030,23 +6517,24 @@ inline bool Server::write_response_core(Stream &strm, bool close_connection, if (close_connection || req.get_header_value("Connection") == "close") { res.set_header("Connection", "close"); } else { - std::stringstream ss; - ss << "timeout=" << keep_alive_timeout_sec_ - << ", max=" << keep_alive_max_count_; - res.set_header("Keep-Alive", ss.str()); + std::string s = "timeout="; + s += std::to_string(keep_alive_timeout_sec_); + s += ", max="; + s += std::to_string(keep_alive_max_count_); + res.set_header("Keep-Alive", s); } - if (!res.has_header("Content-Type") && - (!res.body.empty() || res.content_length_ > 0 || res.content_provider_)) { + if ((!res.body.empty() || res.content_length_ > 0 || res.content_provider_) && + !res.has_header("Content-Type")) { res.set_header("Content-Type", "text/plain"); } - if (!res.has_header("Content-Length") && res.body.empty() && - !res.content_length_ && !res.content_provider_) { + if (res.body.empty() && !res.content_length_ && !res.content_provider_ && + !res.has_header("Content-Length")) { res.set_header("Content-Length", "0"); } - if (!res.has_header("Accept-Ranges") && req.method == "HEAD") { + if (req.method == "HEAD" && !res.has_header("Accept-Ranges")) { res.set_header("Accept-Ranges", "bytes"); } @@ -6055,12 +6543,7 @@ inline bool Server::write_response_core(Stream &strm, bool close_connection, // Response line and headers { detail::BufferStream bstrm; - - if (!bstrm.write_format("HTTP/1.1 %d %s\r\n", res.status, - status_message(res.status))) { - return false; - } - + if (!detail::write_response_line(bstrm, res.status)) { return false; } if (!header_writer_(bstrm, res.headers)) { return false; } // Flush buffer @@ -6254,7 +6737,14 @@ inline bool Server::handle_file_request(const Request &req, Response &res, auto path = entry.base_dir + sub_path; if (path.back() == '/') { path += "index.html"; } - if (detail::is_file(path)) { + detail::FileStat stat(path); + + if (stat.is_dir()) { + res.set_redirect(sub_path + "/", StatusCode::MovedPermanently_301); + return true; + } + + if (stat.is_file()) { for (const auto &kv : entry.headers) { res.set_header(kv.first, kv.second); } @@ -6289,8 +6779,8 @@ Server::create_server_socket(const std::string &host, int port, SocketOptions socket_options) const { return detail::create_socket( host, std::string(), port, address_family_, socket_flags, tcp_nodelay_, - std::move(socket_options), - [](socket_t sock, struct addrinfo &ai) -> bool { + ipv6_v6only_, std::move(socket_options), + [](socket_t sock, struct addrinfo &ai, bool & /*quit*/) -> bool { if (::bind(sock, ai.ai_addr, static_cast(ai.ai_addrlen))) { return false; } @@ -6301,6 +6791,8 @@ Server::create_server_socket(const std::string &host, int port, inline int Server::bind_internal(const std::string &host, int port, int socket_flags) { + if (is_decommisioned) { return -1; } + if (!is_valid()) { return -1; } svr_sock_ = create_server_socket(host, port, socket_flags, socket_options_); @@ -6326,6 +6818,8 @@ inline int Server::bind_internal(const std::string &host, int port, } inline bool Server::listen_internal() { + if (is_decommisioned) { return false; } + auto ret = true; is_running_ = true; auto se = detail::scope_exit([&]() { is_running_ = false; }); @@ -6346,13 +6840,22 @@ inline bool Server::listen_internal() { #ifndef _WIN32 } #endif + +#if defined _WIN32 + // sockets conneced via WASAccept inherit flags NO_HANDLE_INHERIT, + // OVERLAPPED + socket_t sock = WSAAccept(svr_sock_, nullptr, nullptr, nullptr, 0); +#elif defined SOCK_CLOEXEC + socket_t sock = accept4(svr_sock_, nullptr, nullptr, SOCK_CLOEXEC); +#else socket_t sock = accept(svr_sock_, nullptr, nullptr); +#endif if (sock == INVALID_SOCKET) { if (errno == EMFILE) { // The per-process limit of open file descriptors has been reached. // Try to accept new connections after a short sleep. - std::this_thread::sleep_for(std::chrono::milliseconds(1)); + std::this_thread::sleep_for(std::chrono::microseconds{1}); continue; } else if (errno == EINTR || errno == EAGAIN) { continue; @@ -6406,6 +6909,7 @@ inline bool Server::listen_internal() { task_queue->shutdown(); } + is_decommisioned = !ret; return ret; } @@ -6503,7 +7007,7 @@ inline bool Server::dispatch_request(Request &req, Response &res, inline void Server::apply_ranges(const Request &req, Response &res, std::string &content_type, std::string &boundary) const { - if (req.ranges.size() > 1) { + if (req.ranges.size() > 1 && res.status == StatusCode::PartialContent_206) { auto it = res.headers.find("Content-Type"); if (it != res.headers.end()) { content_type = it->second; @@ -6521,7 +7025,7 @@ inline void Server::apply_ranges(const Request &req, Response &res, if (res.body.empty()) { if (res.content_length_ > 0) { size_t length = 0; - if (req.ranges.empty()) { + if (req.ranges.empty() || res.status != StatusCode::PartialContent_206) { length = res.content_length_; } else if (req.ranges.size() == 1) { auto offset_and_length = detail::get_range_offset_and_length( @@ -6550,7 +7054,7 @@ inline void Server::apply_ranges(const Request &req, Response &res, } } } else { - if (req.ranges.empty()) { + if (req.ranges.empty() || res.status != StatusCode::PartialContent_206) { ; } else if (req.ranges.size() == 1) { auto offset_and_length = @@ -6621,7 +7125,9 @@ inline bool Server::dispatch_request_for_content_reader( } inline bool -Server::process_request(Stream &strm, bool close_connection, +Server::process_request(Stream &strm, const std::string &remote_addr, + int remote_port, const std::string &local_addr, + int local_port, bool close_connection, bool &connection_closed, const std::function &setup_request) { std::array buf{}; @@ -6675,11 +7181,13 @@ Server::process_request(Stream &strm, bool close_connection, connection_closed = true; } - strm.get_remote_ip_and_port(req.remote_addr, req.remote_port); + req.remote_addr = remote_addr; + req.remote_port = remote_port; req.set_header("REMOTE_ADDR", req.remote_addr); req.set_header("REMOTE_PORT", std::to_string(req.remote_port)); - strm.get_local_ip_and_port(req.local_addr, req.local_port); + req.local_addr = local_addr; + req.local_port = local_port; req.set_header("LOCAL_ADDR", req.local_addr); req.set_header("LOCAL_PORT", std::to_string(req.local_port)); @@ -6701,13 +7209,20 @@ Server::process_request(Stream &strm, bool close_connection, switch (status) { case StatusCode::Continue_100: case StatusCode::ExpectationFailed_417: - strm.write_format("HTTP/1.1 %d %s\r\n\r\n", status, - status_message(status)); + detail::write_response_line(strm, status); + strm.write("\r\n"); break; - default: return write_response(strm, close_connection, req, res); + default: + connection_closed = true; + return write_response(strm, true, req, res); } } + // Setup `is_connection_closed` method + req.is_connection_closed = [&]() { + return !detail::is_socket_alive(strm.socket()); + }; + // Routing auto routed = false; #ifdef CPPHTTPLIB_NO_EXCEPTIONS @@ -6750,6 +7265,32 @@ Server::process_request(Stream &strm, bool close_connection, : StatusCode::PartialContent_206; } + // Serve file content by using a content provider + if (!res.file_content_path_.empty()) { + const auto &path = res.file_content_path_; + auto mm = std::make_shared(path.c_str()); + if (!mm->is_open()) { + res.body.clear(); + res.content_length_ = 0; + res.content_provider_ = nullptr; + res.status = StatusCode::NotFound_404; + return write_response(strm, close_connection, req, res); + } + + auto content_type = res.file_content_content_type_; + if (content_type.empty()) { + content_type = detail::find_content_type( + path, file_extension_and_mimetype_map_, default_file_mimetype_); + } + + res.set_content_provider( + mm->size(), content_type, + [mm](size_t offset, size_t length, DataSink &sink) -> bool { + sink.write(mm->data() + offset, length); + return true; + }); + } + if (detail::range_error(req, res)) { res.body.clear(); res.content_length_ = 0; @@ -6769,12 +7310,21 @@ Server::process_request(Stream &strm, bool close_connection, inline bool Server::is_valid() const { return true; } inline bool Server::process_and_close_socket(socket_t sock) { + std::string remote_addr; + int remote_port = 0; + detail::get_remote_ip_and_port(sock, remote_addr, remote_port); + + std::string local_addr; + int local_port = 0; + detail::get_local_ip_and_port(sock, local_addr, local_port); + auto ret = detail::process_server_socket( svr_sock_, sock, keep_alive_max_count_, keep_alive_timeout_sec_, read_timeout_sec_, read_timeout_usec_, write_timeout_sec_, write_timeout_usec_, - [this](Stream &strm, bool close_connection, bool &connection_closed) { - return process_request(strm, close_connection, connection_closed, + [&](Stream &strm, bool close_connection, bool &connection_closed) { + return process_request(strm, remote_addr, remote_port, local_addr, + local_port, close_connection, connection_closed, nullptr); }); @@ -6793,8 +7343,8 @@ inline ClientImpl::ClientImpl(const std::string &host, int port) inline ClientImpl::ClientImpl(const std::string &host, int port, const std::string &client_cert_path, const std::string &client_key_path) - : host_(host), port_(port), - host_and_port_(adjust_host_string(host) + ":" + std::to_string(port)), + : host_(detail::escape_abstract_namespace_unix_domain(host)), port_(port), + host_and_port_(adjust_host_string(host_) + ":" + std::to_string(port)), client_cert_path_(client_cert_path), client_key_path_(client_key_path) {} inline ClientImpl::~ClientImpl() { @@ -6825,6 +7375,7 @@ inline void ClientImpl::copy_settings(const ClientImpl &rhs) { url_encode_ = rhs.url_encode_; address_family_ = rhs.address_family_; tcp_nodelay_ = rhs.tcp_nodelay_; + ipv6_v6only_ = rhs.ipv6_v6only_; socket_options_ = rhs.socket_options_; compress_ = rhs.compress_; decompress_ = rhs.decompress_; @@ -6845,6 +7396,8 @@ inline void ClientImpl::copy_settings(const ClientImpl &rhs) { #endif #ifdef CPPHTTPLIB_OPENSSL_SUPPORT server_certificate_verification_ = rhs.server_certificate_verification_; + server_hostname_verification_ = rhs.server_hostname_verification_; + server_certificate_verifier_ = rhs.server_certificate_verifier_; #endif logger_ = rhs.logger_; } @@ -6853,9 +7406,9 @@ inline socket_t ClientImpl::create_client_socket(Error &error) const { if (!proxy_host_.empty() && proxy_port_ != -1) { return detail::create_client_socket( proxy_host_, std::string(), proxy_port_, address_family_, tcp_nodelay_, - socket_options_, connection_timeout_sec_, connection_timeout_usec_, - read_timeout_sec_, read_timeout_usec_, write_timeout_sec_, - write_timeout_usec_, interface_, error); + ipv6_v6only_, socket_options_, connection_timeout_sec_, + connection_timeout_usec_, read_timeout_sec_, read_timeout_usec_, + write_timeout_sec_, write_timeout_usec_, interface_, error); } // Check is custom IP specified for host_ @@ -6864,10 +7417,10 @@ inline socket_t ClientImpl::create_client_socket(Error &error) const { if (it != addr_map_.end()) { ip = it->second; } return detail::create_client_socket( - host_, ip, port_, address_family_, tcp_nodelay_, socket_options_, - connection_timeout_sec_, connection_timeout_usec_, read_timeout_sec_, - read_timeout_usec_, write_timeout_sec_, write_timeout_usec_, interface_, - error); + host_, ip, port_, address_family_, tcp_nodelay_, ipv6_v6only_, + socket_options_, connection_timeout_sec_, connection_timeout_usec_, + read_timeout_sec_, read_timeout_usec_, write_timeout_sec_, + write_timeout_usec_, interface_, error); } inline bool ClientImpl::create_and_connect_socket(Socket &socket, @@ -6956,6 +7509,18 @@ inline bool ClientImpl::send(Request &req, Response &res, Error &error) { return ret; } +#ifdef CPPHTTPLIB_OPENSSL_SUPPORT +inline bool ClientImpl::is_ssl_peer_could_be_closed(SSL *ssl) const { + detail::set_nonblocking(socket_.sock, true); + auto se = detail::scope_exit( + [&]() { detail::set_nonblocking(socket_.sock, false); }); + + char buf[1]; + return !SSL_peek(ssl, buf, 1) && + SSL_get_error(ssl, 0) == SSL_ERROR_ZERO_RETURN; +} +#endif + inline bool ClientImpl::send_(Request &req, Response &res, Error &error) { { std::lock_guard guard(socket_mutex_); @@ -6967,6 +7532,13 @@ inline bool ClientImpl::send_(Request &req, Response &res, Error &error) { auto is_alive = false; if (socket_.is_open()) { is_alive = detail::is_socket_alive(socket_.sock); + +#ifdef CPPHTTPLIB_OPENSSL_SUPPORT + if (is_alive && is_ssl()) { + if (is_ssl_peer_could_be_closed(socket_.ssl)) { is_alive = false; } + } +#endif + if (!is_alive) { // Attempt to avoid sigpipe by shutting down nongracefully if it seems // like the other side has already closed the connection Also, there @@ -7144,7 +7716,7 @@ inline bool ClientImpl::redirect(Request &req, Response &res, Error &error) { if (location.empty()) { return false; } const static std::regex re( - R"((?:(https?):)?(?://(?:\[([\d:]+)\]|([^:/?#]+))(?::(\d+))?)?([^?#]*)(\?[^#]*)?(?:#.*)?)"); + R"((?:(https?):)?(?://(?:\[([a-fA-F\d:]+)\]|([^:/?#]+))(?::(\d+))?)?([^?#]*)(\?[^#]*)?(?:#.*)?)"); std::smatch m; if (!std::regex_match(location, m, re)) { return false; } @@ -7243,12 +7815,26 @@ inline bool ClientImpl::write_request(Stream &strm, Request &req, if (!req.has_header("Accept")) { req.set_header("Accept", "*/*"); } -#ifndef CPPHTTPLIB_NO_DEFAULT_USER_AGENT - if (!req.has_header("User-Agent")) { - auto agent = std::string("cpp-httplib/") + CPPHTTPLIB_VERSION; - req.set_header("User-Agent", agent); - } + if (!req.content_receiver) { + if (!req.has_header("Accept-Encoding")) { + std::string accept_encoding; +#ifdef CPPHTTPLIB_BROTLI_SUPPORT + accept_encoding = "br"; #endif +#ifdef CPPHTTPLIB_ZLIB_SUPPORT + if (!accept_encoding.empty()) { accept_encoding += ", "; } + accept_encoding += "gzip, deflate"; +#endif + req.set_header("Accept-Encoding", accept_encoding); + } + +#ifndef CPPHTTPLIB_NO_DEFAULT_USER_AGENT + if (!req.has_header("User-Agent")) { + auto agent = std::string("cpp-httplib/") + CPPHTTPLIB_VERSION; + req.set_header("User-Agent", agent); + } +#endif + }; if (req.body.empty()) { if (req.content_provider_) { @@ -7308,8 +7894,14 @@ inline bool ClientImpl::write_request(Stream &strm, Request &req, { detail::BufferStream bstrm; - const auto &path = url_encode_ ? detail::encode_url(req.path) : req.path; - bstrm.write_format("%s %s HTTP/1.1\r\n", req.method.c_str(), path.c_str()); + const auto &path_with_query = + req.params.empty() ? req.path + : append_query_params(req.path, req.params); + + const auto &path = + url_encode_ ? detail::encode_url(path_with_query) : path_with_query; + + detail::write_request_line(bstrm, req.method, path); header_writer_(bstrm, req.headers); @@ -7417,11 +8009,12 @@ inline Result ClientImpl::send_with_content_provider( const std::string &method, const std::string &path, const Headers &headers, const char *body, size_t content_length, ContentProvider content_provider, ContentProviderWithoutLength content_provider_without_length, - const std::string &content_type) { + const std::string &content_type, Progress progress) { Request req; req.method = method; req.headers = headers; req.path = path; + req.progress = progress; auto error = Error::Success; @@ -7448,9 +8041,7 @@ inline bool ClientImpl::process_request(Stream &strm, Request &req, if (is_ssl()) { auto is_proxy_enabled = !proxy_host_.empty() && proxy_port_ != -1; if (!is_proxy_enabled) { - char buf[1]; - if (SSL_peek(socket_.ssl, buf, 1) == 0 && - SSL_get_error(socket_.ssl, 0) == SSL_ERROR_ZERO_RETURN) { + if (is_ssl_peer_could_be_closed(socket_.ssl)) { error = Error::SSLPeerCouldBeClosed_; return false; } @@ -7468,7 +8059,9 @@ inline bool ClientImpl::process_request(Stream &strm, Request &req, // Body if ((res.status != StatusCode::NoContent_204) && req.method != "HEAD" && req.method != "CONNECT") { - auto redirect = 300 < res.status && res.status < 400 && follow_location_; + auto redirect = 300 < res.status && res.status < 400 && + res.status != StatusCode::NotModified_304 && + follow_location_; if (req.response_handler && !redirect) { if (!req.response_handler(res)) { @@ -7489,9 +8082,7 @@ inline bool ClientImpl::process_request(Stream &strm, Request &req, : static_cast( [&](const char *buf, size_t n, uint64_t /*off*/, uint64_t /*len*/) { - if (res.body.size() + n > res.body.max_size()) { - return false; - } + assert(res.body.size() + n <= res.body.max_size()); res.body.append(buf, n); return true; }); @@ -7503,12 +8094,25 @@ inline bool ClientImpl::process_request(Stream &strm, Request &req, return ret; }; - int dummy_status; - if (!detail::read_content(strm, res, (std::numeric_limits::max)(), - dummy_status, std::move(progress), std::move(out), - decompress_)) { - if (error != Error::Canceled) { error = Error::Read; } - return false; + if (res.has_header("Content-Length")) { + if (!req.content_receiver) { + auto len = res.get_header_value_u64("Content-Length"); + if (len > res.body.max_size()) { + error = Error::Read; + return false; + } + res.body.reserve(static_cast(len)); + } + } + + if (res.status != StatusCode::NotModified_304) { + int dummy_status; + if (!detail::read_content(strm, res, (std::numeric_limits::max)(), + dummy_status, std::move(progress), + std::move(out), decompress_)) { + if (error != Error::Canceled) { error = Error::Read; } + return false; + } } } @@ -7717,14 +8321,22 @@ inline Result ClientImpl::Post(const std::string &path, inline Result ClientImpl::Post(const std::string &path, const char *body, size_t content_length, const std::string &content_type) { - return Post(path, Headers(), body, content_length, content_type); + return Post(path, Headers(), body, content_length, content_type, nullptr); } inline Result ClientImpl::Post(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type) { return send_with_content_provider("POST", path, headers, body, content_length, - nullptr, nullptr, content_type); + nullptr, nullptr, content_type, nullptr); +} + +inline Result ClientImpl::Post(const std::string &path, const Headers &headers, + const char *body, size_t content_length, + const std::string &content_type, + Progress progress) { + return send_with_content_provider("POST", path, headers, body, content_length, + nullptr, nullptr, content_type, progress); } inline Result ClientImpl::Post(const std::string &path, const std::string &body, @@ -7732,12 +8344,27 @@ inline Result ClientImpl::Post(const std::string &path, const std::string &body, return Post(path, Headers(), body, content_type); } +inline Result ClientImpl::Post(const std::string &path, const std::string &body, + const std::string &content_type, + Progress progress) { + return Post(path, Headers(), body, content_type, progress); +} + inline Result ClientImpl::Post(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type) { return send_with_content_provider("POST", path, headers, body.data(), - body.size(), nullptr, nullptr, - content_type); + body.size(), nullptr, nullptr, content_type, + nullptr); +} + +inline Result ClientImpl::Post(const std::string &path, const Headers &headers, + const std::string &body, + const std::string &content_type, + Progress progress) { + return send_with_content_provider("POST", path, headers, body.data(), + body.size(), nullptr, nullptr, content_type, + progress); } inline Result ClientImpl::Post(const std::string &path, const Params ¶ms) { @@ -7763,14 +8390,15 @@ inline Result ClientImpl::Post(const std::string &path, const Headers &headers, const std::string &content_type) { return send_with_content_provider("POST", path, headers, nullptr, content_length, std::move(content_provider), - nullptr, content_type); + nullptr, content_type, nullptr); } inline Result ClientImpl::Post(const std::string &path, const Headers &headers, ContentProviderWithoutLength content_provider, const std::string &content_type) { return send_with_content_provider("POST", path, headers, nullptr, 0, nullptr, - std::move(content_provider), content_type); + std::move(content_provider), content_type, + nullptr); } inline Result ClientImpl::Post(const std::string &path, const Headers &headers, @@ -7779,6 +8407,13 @@ inline Result ClientImpl::Post(const std::string &path, const Headers &headers, return Post(path, headers, query, "application/x-www-form-urlencoded"); } +inline Result ClientImpl::Post(const std::string &path, const Headers &headers, + const Params ¶ms, Progress progress) { + auto query = detail::params_to_query_str(params); + return Post(path, headers, query, "application/x-www-form-urlencoded", + progress); +} + inline Result ClientImpl::Post(const std::string &path, const MultipartFormDataItems &items) { return Post(path, Headers(), items); @@ -7816,7 +8451,7 @@ ClientImpl::Post(const std::string &path, const Headers &headers, return send_with_content_provider( "POST", path, headers, nullptr, 0, nullptr, get_multipart_content_provider(boundary, items, provider_items), - content_type); + content_type, nullptr); } inline Result ClientImpl::Put(const std::string &path) { @@ -7833,7 +8468,15 @@ inline Result ClientImpl::Put(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type) { return send_with_content_provider("PUT", path, headers, body, content_length, - nullptr, nullptr, content_type); + nullptr, nullptr, content_type, nullptr); +} + +inline Result ClientImpl::Put(const std::string &path, const Headers &headers, + const char *body, size_t content_length, + const std::string &content_type, + Progress progress) { + return send_with_content_provider("PUT", path, headers, body, content_length, + nullptr, nullptr, content_type, progress); } inline Result ClientImpl::Put(const std::string &path, const std::string &body, @@ -7841,12 +8484,27 @@ inline Result ClientImpl::Put(const std::string &path, const std::string &body, return Put(path, Headers(), body, content_type); } +inline Result ClientImpl::Put(const std::string &path, const std::string &body, + const std::string &content_type, + Progress progress) { + return Put(path, Headers(), body, content_type, progress); +} + inline Result ClientImpl::Put(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type) { return send_with_content_provider("PUT", path, headers, body.data(), - body.size(), nullptr, nullptr, - content_type); + body.size(), nullptr, nullptr, content_type, + nullptr); +} + +inline Result ClientImpl::Put(const std::string &path, const Headers &headers, + const std::string &body, + const std::string &content_type, + Progress progress) { + return send_with_content_provider("PUT", path, headers, body.data(), + body.size(), nullptr, nullptr, content_type, + progress); } inline Result ClientImpl::Put(const std::string &path, size_t content_length, @@ -7868,14 +8526,15 @@ inline Result ClientImpl::Put(const std::string &path, const Headers &headers, const std::string &content_type) { return send_with_content_provider("PUT", path, headers, nullptr, content_length, std::move(content_provider), - nullptr, content_type); + nullptr, content_type, nullptr); } inline Result ClientImpl::Put(const std::string &path, const Headers &headers, ContentProviderWithoutLength content_provider, const std::string &content_type) { return send_with_content_provider("PUT", path, headers, nullptr, 0, nullptr, - std::move(content_provider), content_type); + std::move(content_provider), content_type, + nullptr); } inline Result ClientImpl::Put(const std::string &path, const Params ¶ms) { @@ -7888,6 +8547,13 @@ inline Result ClientImpl::Put(const std::string &path, const Headers &headers, return Put(path, headers, query, "application/x-www-form-urlencoded"); } +inline Result ClientImpl::Put(const std::string &path, const Headers &headers, + const Params ¶ms, Progress progress) { + auto query = detail::params_to_query_str(params); + return Put(path, headers, query, "application/x-www-form-urlencoded", + progress); +} + inline Result ClientImpl::Put(const std::string &path, const MultipartFormDataItems &items) { return Put(path, Headers(), items); @@ -7925,7 +8591,7 @@ ClientImpl::Put(const std::string &path, const Headers &headers, return send_with_content_provider( "PUT", path, headers, nullptr, 0, nullptr, get_multipart_content_provider(boundary, items, provider_items), - content_type); + content_type, nullptr); } inline Result ClientImpl::Patch(const std::string &path) { return Patch(path, std::string(), std::string()); @@ -7937,12 +8603,26 @@ inline Result ClientImpl::Patch(const std::string &path, const char *body, return Patch(path, Headers(), body, content_length, content_type); } +inline Result ClientImpl::Patch(const std::string &path, const char *body, + size_t content_length, + const std::string &content_type, + Progress progress) { + return Patch(path, Headers(), body, content_length, content_type, progress); +} + inline Result ClientImpl::Patch(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type) { + return Patch(path, headers, body, content_length, content_type, nullptr); +} + +inline Result ClientImpl::Patch(const std::string &path, const Headers &headers, + const char *body, size_t content_length, + const std::string &content_type, + Progress progress) { return send_with_content_provider("PATCH", path, headers, body, content_length, nullptr, nullptr, - content_type); + content_type, progress); } inline Result ClientImpl::Patch(const std::string &path, @@ -7951,12 +8631,26 @@ inline Result ClientImpl::Patch(const std::string &path, return Patch(path, Headers(), body, content_type); } +inline Result ClientImpl::Patch(const std::string &path, + const std::string &body, + const std::string &content_type, + Progress progress) { + return Patch(path, Headers(), body, content_type, progress); +} + inline Result ClientImpl::Patch(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type) { + return Patch(path, headers, body, content_type, nullptr); +} + +inline Result ClientImpl::Patch(const std::string &path, const Headers &headers, + const std::string &body, + const std::string &content_type, + Progress progress) { return send_with_content_provider("PATCH", path, headers, body.data(), - body.size(), nullptr, nullptr, - content_type); + body.size(), nullptr, nullptr, content_type, + progress); } inline Result ClientImpl::Patch(const std::string &path, size_t content_length, @@ -7978,14 +8672,15 @@ inline Result ClientImpl::Patch(const std::string &path, const Headers &headers, const std::string &content_type) { return send_with_content_provider("PATCH", path, headers, nullptr, content_length, std::move(content_provider), - nullptr, content_type); + nullptr, content_type, nullptr); } inline Result ClientImpl::Patch(const std::string &path, const Headers &headers, ContentProviderWithoutLength content_provider, const std::string &content_type) { return send_with_content_provider("PATCH", path, headers, nullptr, 0, nullptr, - std::move(content_provider), content_type); + std::move(content_provider), content_type, + nullptr); } inline Result ClientImpl::Delete(const std::string &path) { @@ -8003,14 +8698,30 @@ inline Result ClientImpl::Delete(const std::string &path, const char *body, return Delete(path, Headers(), body, content_length, content_type); } +inline Result ClientImpl::Delete(const std::string &path, const char *body, + size_t content_length, + const std::string &content_type, + Progress progress) { + return Delete(path, Headers(), body, content_length, content_type, progress); +} + inline Result ClientImpl::Delete(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type) { + return Delete(path, headers, body, content_length, content_type, nullptr); +} + +inline Result ClientImpl::Delete(const std::string &path, + const Headers &headers, const char *body, + size_t content_length, + const std::string &content_type, + Progress progress) { Request req; req.method = "DELETE"; req.headers = headers; req.path = path; + req.progress = progress; if (!content_type.empty()) { req.set_header("Content-Type", content_type); } req.body.assign(body, content_length); @@ -8024,6 +8735,14 @@ inline Result ClientImpl::Delete(const std::string &path, return Delete(path, Headers(), body.data(), body.size(), content_type); } +inline Result ClientImpl::Delete(const std::string &path, + const std::string &body, + const std::string &content_type, + Progress progress) { + return Delete(path, Headers(), body.data(), body.size(), content_type, + progress); +} + inline Result ClientImpl::Delete(const std::string &path, const Headers &headers, const std::string &body, @@ -8031,6 +8750,15 @@ inline Result ClientImpl::Delete(const std::string &path, return Delete(path, headers, body.data(), body.size(), content_type); } +inline Result ClientImpl::Delete(const std::string &path, + const Headers &headers, + const std::string &body, + const std::string &content_type, + Progress progress) { + return Delete(path, headers, body.data(), body.size(), content_type, + progress); +} + inline Result ClientImpl::Options(const std::string &path) { return Options(path, Headers()); } @@ -8138,6 +8866,8 @@ inline void ClientImpl::set_address_family(int family) { inline void ClientImpl::set_tcp_nodelay(bool on) { tcp_nodelay_ = on; } +inline void ClientImpl::set_ipv6_v6only(bool on) { ipv6_v6only_ = on; } + inline void ClientImpl::set_socket_options(SocketOptions socket_options) { socket_options_ = std::move(socket_options); } @@ -8187,13 +8917,11 @@ inline void ClientImpl::set_ca_cert_store(X509_STORE *ca_cert_store) { inline X509_STORE *ClientImpl::create_ca_cert_store(const char *ca_cert, std::size_t size) const { auto mem = BIO_new_mem_buf(ca_cert, static_cast(size)); + auto se = detail::scope_exit([&] { BIO_free_all(mem); }); if (!mem) { return nullptr; } auto inf = PEM_X509_INFO_read_bio(mem, nullptr, nullptr, nullptr); - if (!inf) { - BIO_free_all(mem); - return nullptr; - } + if (!inf) { return nullptr; } auto cts = X509_STORE_new(); if (cts) { @@ -8207,13 +8935,21 @@ inline X509_STORE *ClientImpl::create_ca_cert_store(const char *ca_cert, } sk_X509_INFO_pop_free(inf, X509_INFO_free); - BIO_free_all(mem); return cts; } inline void ClientImpl::enable_server_certificate_verification(bool enabled) { server_certificate_verification_ = enabled; } + +inline void ClientImpl::enable_server_hostname_verification(bool enabled) { + server_hostname_verification_ = enabled; +} + +inline void ClientImpl::set_server_certificate_verifier( + std::function verifier) { + server_certificate_verifier_ = verifier; +} #endif inline void ClientImpl::set_logger(Logger logger) { @@ -8257,13 +8993,30 @@ inline SSL *ssl_new(socket_t sock, SSL_CTX *ctx, std::mutex &ctx_mutex, return ssl; } -inline void ssl_delete(std::mutex &ctx_mutex, SSL *ssl, +inline void ssl_delete(std::mutex &ctx_mutex, SSL *ssl, socket_t sock, bool shutdown_gracefully) { // sometimes we may want to skip this to try to avoid SIGPIPE if we know // the remote has closed the network connection // Note that it is not always possible to avoid SIGPIPE, this is merely a // best-efforts. - if (shutdown_gracefully) { SSL_shutdown(ssl); } + if (shutdown_gracefully) { +#ifdef _WIN32 + (void)(sock); + SSL_shutdown(ssl); +#else + timeval tv; + tv.tv_sec = 1; + tv.tv_usec = 0; + setsockopt(sock, SOL_SOCKET, SO_RCVTIMEO, + reinterpret_cast(&tv), sizeof(tv)); + + auto ret = SSL_shutdown(ssl); + while (ret == 0) { + std::this_thread::sleep_for(std::chrono::milliseconds{100}); + ret = SSL_shutdown(ssl); + } +#endif + } std::lock_guard guard(ctx_mutex); SSL_free(ssl); @@ -8366,7 +9119,7 @@ inline ssize_t SSLSocketStream::read(char *ptr, size_t size) { if (SSL_pending(ssl_) > 0) { return SSL_read(ssl_, ptr, static_cast(size)); } else if (is_readable()) { - std::this_thread::sleep_for(std::chrono::milliseconds(1)); + std::this_thread::sleep_for(std::chrono::microseconds{10}); ret = SSL_read(ssl_, ptr, static_cast(size)); if (ret >= 0) { return ret; } err = SSL_get_error(ssl_, ret); @@ -8397,7 +9150,7 @@ inline ssize_t SSLSocketStream::write(const char *ptr, size_t size) { while (--n >= 0 && err == SSL_ERROR_WANT_WRITE) { #endif if (is_writable()) { - std::this_thread::sleep_for(std::chrono::milliseconds(1)); + std::this_thread::sleep_for(std::chrono::microseconds{10}); ret = SSL_write(ssl_, ptr, static_cast(handle_size)); if (ret >= 0) { return ret; } err = SSL_get_error(ssl_, ret); @@ -8439,7 +9192,7 @@ inline SSLServer::SSLServer(const char *cert_path, const char *private_key_path, SSL_OP_NO_COMPRESSION | SSL_OP_NO_SESSION_RESUMPTION_ON_RENEGOTIATION); - SSL_CTX_set_min_proto_version(ctx_, TLS1_1_VERSION); + SSL_CTX_set_min_proto_version(ctx_, TLS1_2_VERSION); if (private_key_password != nullptr && (private_key_password[0] != '\0')) { SSL_CTX_set_default_passwd_cb_userdata( @@ -8449,7 +9202,8 @@ inline SSLServer::SSLServer(const char *cert_path, const char *private_key_path, if (SSL_CTX_use_certificate_chain_file(ctx_, cert_path) != 1 || SSL_CTX_use_PrivateKey_file(ctx_, private_key_path, SSL_FILETYPE_PEM) != - 1) { + 1 || + SSL_CTX_check_private_key(ctx_) != 1) { SSL_CTX_free(ctx_); ctx_ = nullptr; } else if (client_ca_cert_file_path || client_ca_cert_dir_path) { @@ -8471,7 +9225,7 @@ inline SSLServer::SSLServer(X509 *cert, EVP_PKEY *private_key, SSL_OP_NO_COMPRESSION | SSL_OP_NO_SESSION_RESUMPTION_ON_RENEGOTIATION); - SSL_CTX_set_min_proto_version(ctx_, TLS1_1_VERSION); + SSL_CTX_set_min_proto_version(ctx_, TLS1_2_VERSION); if (SSL_CTX_use_certificate(ctx_, cert) != 1 || SSL_CTX_use_PrivateKey(ctx_, private_key) != 1) { @@ -8505,6 +9259,19 @@ inline bool SSLServer::is_valid() const { return ctx_; } inline SSL_CTX *SSLServer::ssl_context() const { return ctx_; } +inline void SSLServer::update_certs(X509 *cert, EVP_PKEY *private_key, + X509_STORE *client_ca_cert_store) { + + std::lock_guard guard(ctx_mutex_); + + SSL_CTX_use_certificate(ctx_, cert); + SSL_CTX_use_PrivateKey(ctx_, private_key); + + if (client_ca_cert_store != nullptr) { + SSL_CTX_set_cert_store(ctx_, client_ca_cert_store); + } +} + inline bool SSLServer::process_and_close_socket(socket_t sock) { auto ssl = detail::ssl_new( sock, ctx_, ctx_mutex_, @@ -8516,20 +9283,29 @@ inline bool SSLServer::process_and_close_socket(socket_t sock) { auto ret = false; if (ssl) { + std::string remote_addr; + int remote_port = 0; + detail::get_remote_ip_and_port(sock, remote_addr, remote_port); + + std::string local_addr; + int local_port = 0; + detail::get_local_ip_and_port(sock, local_addr, local_port); + ret = detail::process_server_socket_ssl( svr_sock_, ssl, sock, keep_alive_max_count_, keep_alive_timeout_sec_, read_timeout_sec_, read_timeout_usec_, write_timeout_sec_, write_timeout_usec_, - [this, ssl](Stream &strm, bool close_connection, - bool &connection_closed) { - return process_request(strm, close_connection, connection_closed, + [&](Stream &strm, bool close_connection, bool &connection_closed) { + return process_request(strm, remote_addr, remote_port, local_addr, + local_port, close_connection, + connection_closed, [&](Request &req) { req.ssl = ssl; }); }); // Shutdown gracefully if the result seemed successful, non-gracefully if // the connection appeared to be closed. const bool shutdown_gracefully = ret; - detail::ssl_delete(ctx_mutex_, ssl, shutdown_gracefully); + detail::ssl_delete(ctx_mutex_, ssl, sock, shutdown_gracefully); } detail::shutdown_socket(sock); @@ -8551,6 +9327,8 @@ inline SSLClient::SSLClient(const std::string &host, int port, : ClientImpl(host, port, client_cert_path, client_key_path) { ctx_ = SSL_CTX_new(TLS_client_method()); + SSL_CTX_set_min_proto_version(ctx_, TLS1_2_VERSION); + detail::split(&host_[0], &host_[host_.size()], '.', [&](const char *b, const char *e) { host_components_.emplace_back(b, e); @@ -8758,36 +9536,47 @@ inline bool SSLClient::initialize_ssl(Socket &socket, Error &error) { } if (server_certificate_verification_) { - verify_result_ = SSL_get_verify_result(ssl2); + if (server_certificate_verifier_) { + if (!server_certificate_verifier_(ssl2)) { + error = Error::SSLServerVerification; + return false; + } + } else { + verify_result_ = SSL_get_verify_result(ssl2); - if (verify_result_ != X509_V_OK) { - error = Error::SSLServerVerification; - return false; + if (verify_result_ != X509_V_OK) { + error = Error::SSLServerVerification; + return false; + } + + auto server_cert = SSL_get1_peer_certificate(ssl2); + auto se = detail::scope_exit([&] { X509_free(server_cert); }); + + if (server_cert == nullptr) { + error = Error::SSLServerVerification; + return false; + } + + if (server_hostname_verification_) { + if (!verify_host(server_cert)) { + error = Error::SSLServerHostnameVerification; + return false; + } + } } - - auto server_cert = SSL_get1_peer_certificate(ssl2); - - if (server_cert == nullptr) { - error = Error::SSLServerVerification; - return false; - } - - if (!verify_host(server_cert)) { - X509_free(server_cert); - error = Error::SSLServerVerification; - return false; - } - X509_free(server_cert); } return true; }, [&](SSL *ssl2) { +#if defined(OPENSSL_IS_BORINGSSL) + SSL_set_tlsext_host_name(ssl2, host_.c_str()); +#else // NOTE: Direct call instead of using the OpenSSL macro to suppress // -Wold-style-cast warning - // SSL_set_tlsext_host_name(ssl2, host_.c_str()); SSL_ctrl(ssl2, SSL_CTRL_SET_TLSEXT_HOSTNAME, TLSEXT_NAMETYPE_host_name, static_cast(const_cast(host_.c_str()))); +#endif return true; }); @@ -8812,7 +9601,8 @@ inline void SSLClient::shutdown_ssl_impl(Socket &socket, return; } if (socket.ssl) { - detail::ssl_delete(ctx_mutex_, socket.ssl, shutdown_gracefully); + detail::ssl_delete(ctx_mutex_, socket.ssl, socket.sock, + shutdown_gracefully); socket.ssl = nullptr; } assert(socket.ssl == nullptr); @@ -8861,8 +9651,8 @@ SSLClient::verify_host_with_subject_alt_name(X509 *server_cert) const { auto type = GEN_DNS; - struct in6_addr addr6 {}; - struct in_addr addr {}; + struct in6_addr addr6{}; + struct in_addr addr{}; size_t addr_len = 0; #ifndef __MINGW32__ @@ -8965,7 +9755,7 @@ inline Client::Client(const std::string &scheme_host_port, const std::string &client_cert_path, const std::string &client_key_path) { const static std::regex re( - R"((?:([a-z]+):\/\/)?(?:\[([\d:]+)\]|([^:/?#]+))(?::(\d+))?)"); + R"((?:([a-z]+):\/\/)?(?:\[([a-fA-F\d:]+)\]|([^:/?#]+))(?::(\d+))?)"); std::smatch m; if (std::regex_match(scheme_host_port, m, re)) { @@ -9002,10 +9792,12 @@ inline Client::Client(const std::string &scheme_host_port, client_key_path); } } else { + // NOTE: Update TEST(UniversalClientImplTest, Ipv6LiteralAddress) + // if port param below changes. cli_ = detail::make_unique(scheme_host_port, 80, client_cert_path, client_key_path); } -} +} // namespace detail inline Client::Client(const std::string &host, int port) : cli_(detail::make_unique(host, port)) {} @@ -9111,15 +9903,30 @@ inline Result Client::Post(const std::string &path, const Headers &headers, const std::string &content_type) { return cli_->Post(path, headers, body, content_length, content_type); } +inline Result Client::Post(const std::string &path, const Headers &headers, + const char *body, size_t content_length, + const std::string &content_type, Progress progress) { + return cli_->Post(path, headers, body, content_length, content_type, + progress); +} inline Result Client::Post(const std::string &path, const std::string &body, const std::string &content_type) { return cli_->Post(path, body, content_type); } +inline Result Client::Post(const std::string &path, const std::string &body, + const std::string &content_type, Progress progress) { + return cli_->Post(path, body, content_type, progress); +} inline Result Client::Post(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type) { return cli_->Post(path, headers, body, content_type); } +inline Result Client::Post(const std::string &path, const Headers &headers, + const std::string &body, + const std::string &content_type, Progress progress) { + return cli_->Post(path, headers, body, content_type, progress); +} inline Result Client::Post(const std::string &path, size_t content_length, ContentProvider content_provider, const std::string &content_type) { @@ -9150,6 +9957,10 @@ inline Result Client::Post(const std::string &path, const Headers &headers, const Params ¶ms) { return cli_->Post(path, headers, params); } +inline Result Client::Post(const std::string &path, const Headers &headers, + const Params ¶ms, Progress progress) { + return cli_->Post(path, headers, params, progress); +} inline Result Client::Post(const std::string &path, const MultipartFormDataItems &items) { return cli_->Post(path, items); @@ -9180,15 +9991,29 @@ inline Result Client::Put(const std::string &path, const Headers &headers, const std::string &content_type) { return cli_->Put(path, headers, body, content_length, content_type); } +inline Result Client::Put(const std::string &path, const Headers &headers, + const char *body, size_t content_length, + const std::string &content_type, Progress progress) { + return cli_->Put(path, headers, body, content_length, content_type, progress); +} inline Result Client::Put(const std::string &path, const std::string &body, const std::string &content_type) { return cli_->Put(path, body, content_type); } +inline Result Client::Put(const std::string &path, const std::string &body, + const std::string &content_type, Progress progress) { + return cli_->Put(path, body, content_type, progress); +} inline Result Client::Put(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type) { return cli_->Put(path, headers, body, content_type); } +inline Result Client::Put(const std::string &path, const Headers &headers, + const std::string &body, + const std::string &content_type, Progress progress) { + return cli_->Put(path, headers, body, content_type, progress); +} inline Result Client::Put(const std::string &path, size_t content_length, ContentProvider content_provider, const std::string &content_type) { @@ -9219,6 +10044,10 @@ inline Result Client::Put(const std::string &path, const Headers &headers, const Params ¶ms) { return cli_->Put(path, headers, params); } +inline Result Client::Put(const std::string &path, const Headers &headers, + const Params ¶ms, Progress progress) { + return cli_->Put(path, headers, params, progress); +} inline Result Client::Put(const std::string &path, const MultipartFormDataItems &items) { return cli_->Put(path, items); @@ -9246,20 +10075,44 @@ inline Result Client::Patch(const std::string &path, const char *body, const std::string &content_type) { return cli_->Patch(path, body, content_length, content_type); } +inline Result Client::Patch(const std::string &path, const char *body, + size_t content_length, + const std::string &content_type, + Progress progress) { + return cli_->Patch(path, body, content_length, content_type, progress); +} inline Result Client::Patch(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type) { return cli_->Patch(path, headers, body, content_length, content_type); } +inline Result Client::Patch(const std::string &path, const Headers &headers, + const char *body, size_t content_length, + const std::string &content_type, + Progress progress) { + return cli_->Patch(path, headers, body, content_length, content_type, + progress); +} inline Result Client::Patch(const std::string &path, const std::string &body, const std::string &content_type) { return cli_->Patch(path, body, content_type); } +inline Result Client::Patch(const std::string &path, const std::string &body, + const std::string &content_type, + Progress progress) { + return cli_->Patch(path, body, content_type, progress); +} inline Result Client::Patch(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type) { return cli_->Patch(path, headers, body, content_type); } +inline Result Client::Patch(const std::string &path, const Headers &headers, + const std::string &body, + const std::string &content_type, + Progress progress) { + return cli_->Patch(path, headers, body, content_type, progress); +} inline Result Client::Patch(const std::string &path, size_t content_length, ContentProvider content_provider, const std::string &content_type) { @@ -9294,20 +10147,44 @@ inline Result Client::Delete(const std::string &path, const char *body, const std::string &content_type) { return cli_->Delete(path, body, content_length, content_type); } +inline Result Client::Delete(const std::string &path, const char *body, + size_t content_length, + const std::string &content_type, + Progress progress) { + return cli_->Delete(path, body, content_length, content_type, progress); +} inline Result Client::Delete(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type) { return cli_->Delete(path, headers, body, content_length, content_type); } +inline Result Client::Delete(const std::string &path, const Headers &headers, + const char *body, size_t content_length, + const std::string &content_type, + Progress progress) { + return cli_->Delete(path, headers, body, content_length, content_type, + progress); +} inline Result Client::Delete(const std::string &path, const std::string &body, const std::string &content_type) { return cli_->Delete(path, body, content_type); } +inline Result Client::Delete(const std::string &path, const std::string &body, + const std::string &content_type, + Progress progress) { + return cli_->Delete(path, body, content_type, progress); +} inline Result Client::Delete(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type) { return cli_->Delete(path, headers, body, content_type); } +inline Result Client::Delete(const std::string &path, const Headers &headers, + const std::string &body, + const std::string &content_type, + Progress progress) { + return cli_->Delete(path, headers, body, content_type, progress); +} inline Result Client::Options(const std::string &path) { return cli_->Options(path); } @@ -9417,6 +10294,15 @@ inline void Client::set_proxy_digest_auth(const std::string &username, inline void Client::enable_server_certificate_verification(bool enabled) { cli_->enable_server_certificate_verification(enabled); } + +inline void Client::enable_server_hostname_verification(bool enabled) { + cli_->enable_server_hostname_verification(enabled); +} + +inline void Client::set_server_certificate_verifier( + std::function verifier) { + cli_->set_server_certificate_verifier(verifier); +} #endif inline void Client::set_logger(Logger logger) { diff --git a/examples/server/public/index.html.gz b/examples/server/public/index.html.gz index 36f9c9fe9..582ccc0d3 100644 Binary files a/examples/server/public/index.html.gz and b/examples/server/public/index.html.gz differ diff --git a/examples/server/server.cpp b/examples/server/server.cpp index c2e62ba69..b1cde2d7f 100644 --- a/examples/server/server.cpp +++ b/examples/server/server.cpp @@ -19,6 +19,7 @@ #include "loading.html.hpp" #include +#include #include #include #include @@ -32,6 +33,8 @@ using json = nlohmann::ordered_json; +constexpr int HTTP_POLLING_SECONDS = 1; + enum stop_type { STOP_TYPE_NONE, STOP_TYPE_EOS, @@ -98,7 +101,7 @@ struct slot_params { int64_t t_max_prompt_ms = -1; // TODO: implement int64_t t_max_predict_ms = -1; // if positive, limit the generation phase to this time limit - std::vector lora; + std::vector lora; std::vector antiprompt; std::vector response_fields; @@ -198,15 +201,17 @@ struct server_task { bool metrics_reset_bucket = false; // used by SERVER_TASK_TYPE_SET_LORA - std::vector set_lora; + std::vector set_lora; server_task(server_task_type type) : type(type) {} static slot_params params_from_json_cmpl( - const llama_model * model, const llama_context * ctx, const common_params & params_base, const json & data) { + const llama_model * model = llama_get_model(ctx); + const llama_vocab * vocab = llama_model_get_vocab(model); + slot_params params; // Sampling parameter defaults are loaded from the global server context (but individual requests can still override them) @@ -262,6 +267,11 @@ struct server_task { params.speculative.n_min = std::max(params.speculative.n_min, 2); params.speculative.n_max = std::max(params.speculative.n_max, 0); + // Use OpenAI API logprobs only if n_probs wasn't provided + if (data.contains("logprobs") && params.sampling.n_probs == defaults.sampling.n_probs){ + params.sampling.n_probs = json_value(data, "logprobs", defaults.sampling.n_probs); + } + if (data.contains("lora")) { if (data.at("lora").is_array()) { params.lora = parse_lora_request(params_base.lora_adapters, data.at("lora")); @@ -329,7 +339,7 @@ struct server_task { const auto & logit_bias = data.find("logit_bias"); if (logit_bias != data.end() && logit_bias->is_array()) { - const int n_vocab = llama_n_vocab(model); + const int n_vocab = llama_vocab_n_tokens(vocab); for (const auto & el : *logit_bias) { // TODO: we may want to throw errors here, in case "el" is incorrect if (el.is_array() && el.size() == 2) { @@ -348,7 +358,7 @@ struct server_task { params.sampling.logit_bias.push_back({tok, bias}); } } else if (el[0].is_string()) { - auto toks = common_tokenize(model, el[0].get(), false); + auto toks = common_tokenize(vocab, el[0].get(), false); for (auto tok : toks) { params.sampling.logit_bias.push_back({tok, bias}); } @@ -1131,7 +1141,7 @@ struct server_slot { common_speculative * spec = nullptr; - std::vector lora; + std::vector lora; // the index relative to completion multi-task request size_t index = 0; @@ -1417,6 +1427,10 @@ struct server_queue { int post(server_task task, bool front = false) { std::unique_lock lock(mutex_tasks); GGML_ASSERT(task.id != -1); + // if this is cancel task make sure to clean up pending tasks + if (task.type == SERVER_TASK_TYPE_CANCEL) { + cleanup_pending_task(task.id_target); + } QUE_DBG("new task, id = %d, front = %d\n", task.id, front); if (front) { queue_tasks.push_front(std::move(task)); @@ -1434,6 +1448,10 @@ struct server_queue { if (task.id == -1) { task.id = id++; } + // if this is cancel task make sure to clean up pending tasks + if (task.type == SERVER_TASK_TYPE_CANCEL) { + cleanup_pending_task(task.id_target); + } QUE_DBG("new task, id = %d/%d, front = %d\n", task.id, (int) tasks.size(), front); if (front) { queue_tasks.push_front(std::move(task)); @@ -1534,6 +1552,20 @@ struct server_queue { } } } + +private: + void cleanup_pending_task(int id_target) { + // no need lock because this is called exclusively by post() + auto rm_func = [id_target](const server_task & task) { + return task.id_target == id_target; + }; + queue_tasks.erase( + std::remove_if(queue_tasks.begin(), queue_tasks.end(), rm_func), + queue_tasks.end()); + queue_tasks_deferred.erase( + std::remove_if(queue_tasks_deferred.begin(), queue_tasks_deferred.end(), rm_func), + queue_tasks_deferred.end()); + } }; struct server_response { @@ -1569,6 +1601,12 @@ struct server_response { std::unique_lock lock(mutex_results); waiting_task_ids.erase(id_task); + // make sure to clean up all pending results + queue_results.erase( + std::remove_if(queue_results.begin(), queue_results.end(), [id_task](const server_task_result_ptr & res) { + return res->id == id_task; + }), + queue_results.end()); } void remove_waiting_task_ids(const std::unordered_set & id_tasks) { @@ -1588,6 +1626,24 @@ struct server_response { return !queue_results.empty(); }); + for (size_t i = 0; i < queue_results.size(); i++) { + if (id_tasks.find(queue_results[i]->id) != id_tasks.end()) { + server_task_result_ptr res = std::move(queue_results[i]); + queue_results.erase(queue_results.begin() + i); + return res; + } + } + } + + // should never reach here + } + + // same as recv(), but have timeout in seconds + // if timeout is reached, nullptr is returned + server_task_result_ptr recv_with_timeout(const std::unordered_set & id_tasks, int timeout) { + while (true) { + std::unique_lock lock(mutex_results); + for (int i = 0; i < (int) queue_results.size(); i++) { if (id_tasks.find(queue_results[i]->id) != id_tasks.end()) { server_task_result_ptr res = std::move(queue_results[i]); @@ -1595,6 +1651,11 @@ struct server_response { return res; } } + + std::cv_status cr_res = condition_results.wait_for(lock, std::chrono::seconds(timeout)); + if (cr_res == std::cv_status::timeout) { + return nullptr; + } } // should never reach here @@ -1633,6 +1694,8 @@ struct server_context { llama_model * model = nullptr; llama_context * ctx = nullptr; + const llama_vocab * vocab = nullptr; + llama_model * model_dft = nullptr; llama_context_params cparams_dft; @@ -1657,6 +1720,8 @@ struct server_context { // Necessary similarity of prompt for slot selection float slot_prompt_similarity = 0.0f; + common_chat_templates chat_templates; + ~server_context() { // Clear any sampling context for (server_slot & slot : slots) { @@ -1690,18 +1755,23 @@ struct server_context { return false; } + vocab = llama_model_get_vocab(model); + n_ctx = llama_n_ctx(ctx); - add_bos_token = llama_add_bos_token(model); - has_eos_token = llama_token_eos(model) != LLAMA_TOKEN_NULL; + add_bos_token = llama_vocab_get_add_bos(vocab); + has_eos_token = llama_vocab_eos(vocab) != LLAMA_TOKEN_NULL; - if (!params_base.speculative.model.empty()) { + if (!params_base.speculative.model.empty() || !params_base.speculative.hf_repo.empty()) { SRV_INF("loading draft model '%s'\n", params_base.speculative.model.c_str()); auto params_dft = params_base; params_dft.devices = params_base.speculative.devices; + params_dft.hf_file = params_base.speculative.hf_file; + params_dft.hf_repo = params_base.speculative.hf_repo; params_dft.model = params_base.speculative.model; + params_dft.model_url = params_base.speculative.model_url; params_dft.n_ctx = params_base.speculative.n_ctx == 0 ? params_base.n_ctx / params_base.n_parallel : params_base.speculative.n_ctx; params_dft.n_gpu_layers = params_base.speculative.n_gpu_layers; params_dft.n_parallel = 1; @@ -1729,15 +1799,44 @@ struct server_context { // force F16 KV cache for the draft model for extra performance cparams_dft.type_k = GGML_TYPE_F16; cparams_dft.type_v = GGML_TYPE_F16; + + // the context is not needed - we will create one for each slot + llama_init_dft.context.reset(); } + chat_templates = common_chat_templates_from_model(model, params_base.chat_template); + GGML_ASSERT(chat_templates.template_default.get() != nullptr); + return true; } - bool validate_builtin_chat_template() const { + bool validate_builtin_chat_template(bool use_jinja) const { llama_chat_message chat[] = {{"user", "test"}}; - int32_t chat_res = llama_chat_apply_template(model, nullptr, chat, 1, true, nullptr, 0); - return chat_res > 0; + + if (use_jinja) { + auto templates = common_chat_templates_from_model(model, ""); + GGML_ASSERT(templates.template_default); + try { + templates.template_default->apply({{ + {"role", "user"}, + {"content", "test"}, + }}, json(), true); + if (templates.template_tool_use) { + templates.template_tool_use->apply({{ + {"role", "user"}, + {"content", "test"}, + }}, json(), true); + } + return true; + } catch (const std::exception & e) { + SRV_ERR("failed to apply template: %s\n", e.what()); + return false; + } + } else { + const char * tmpl = llama_model_chat_template(model, /* name */ nullptr); + const int32_t chat_res = llama_chat_apply_template(tmpl, chat, 1, true, nullptr, 0); + return chat_res > 0; + } } void init() { @@ -1756,7 +1855,7 @@ struct server_context { if (model_dft) { slot.batch_spec = llama_batch_init(params_base.speculative.n_max + 1, 0, 1); - slot.ctx_dft = llama_new_context_with_model(model_dft, cparams_dft); + slot.ctx_dft = llama_init_from_model(model_dft, cparams_dft); if (slot.ctx_dft == nullptr) { SRV_ERR("%s", "failed to create draft context\n"); return; @@ -1891,7 +1990,7 @@ struct server_context { } if (slot.params.ignore_eos && has_eos_token) { - slot.params.sampling.logit_bias.push_back({llama_token_eos(model), -INFINITY}); + slot.params.sampling.logit_bias.push_back({llama_vocab_eos(vocab), -INFINITY}); } { @@ -2047,14 +2146,14 @@ struct server_context { slot.n_decoded, slot.n_prompt_tokens, slot.n_past, slot.n_ctx); } - if (llama_token_is_eog(model, result.tok)) { + if (llama_vocab_is_eog(vocab, result.tok)) { slot.stop = STOP_TYPE_EOS; slot.has_next_token = false; SLT_DBG(slot, "%s", "stopped by EOS\n"); } - const auto n_ctx_train = llama_n_ctx_train(model); + const auto n_ctx_train = llama_model_n_ctx_train(model); if (slot.params.n_predict < 1 && slot.n_predict < 1 && slot.n_prompt_tokens + slot.n_decoded >= n_ctx_train) { slot.truncated = true; @@ -2074,7 +2173,7 @@ struct server_context { void populate_token_probs(const server_slot & slot, completion_token_output & result, bool post_sampling, bool special, int idx) { size_t n_probs = slot.params.sampling.n_probs; - size_t n_vocab = llama_n_vocab(llama_get_model(ctx)); + size_t n_vocab = llama_vocab_n_tokens(vocab); if (post_sampling) { const auto * cur_p = common_sampler_get_candidates(slot.smpl); const size_t max_probs = cur_p->size; @@ -2225,7 +2324,7 @@ struct server_context { res->n_tokens = slot.n_prompt_tokens; res->oaicompat = slot.params.oaicompat; - const int n_embd = llama_n_embd(model); + const int n_embd = llama_model_n_embd(model); std::vector embd_res(n_embd, 0.0f); @@ -2304,8 +2403,8 @@ struct server_context { server_task task(SERVER_TASK_TYPE_CANCEL); task.id_target = id_task; - cancel_tasks.push_back(task); queue_results.remove_waiting_task_id(id_task); + cancel_tasks.push_back(task); } // push to beginning of the queue, so it has highest priority queue_tasks.post(cancel_tasks, true); @@ -2315,10 +2414,21 @@ struct server_context { void receive_multi_results( const std::unordered_set & id_tasks, const std::function&)> & result_handler, - const std::function & error_handler) { + const std::function & error_handler, + const std::function & is_connection_closed) { std::vector results(id_tasks.size()); - for (size_t i = 0; i < id_tasks.size(); i++) { - server_task_result_ptr result = queue_results.recv(id_tasks); + for (int i = 0; i < (int)id_tasks.size(); i++) { + server_task_result_ptr result = queue_results.recv_with_timeout(id_tasks, HTTP_POLLING_SECONDS); + + if (is_connection_closed()) { + cancel_tasks(id_tasks); + return; + } + + if (result == nullptr) { + i--; // retry + continue; + } if (result->is_error()) { error_handler(result->to_json()); @@ -2342,10 +2452,20 @@ struct server_context { void receive_cmpl_results_stream( const std::unordered_set & id_tasks, const std::function & result_handler, - const std::function & error_handler) { + const std::function & error_handler, + const std::function & is_connection_closed) { size_t n_finished = 0; while (true) { - server_task_result_ptr result = queue_results.recv(id_tasks); + server_task_result_ptr result = queue_results.recv_with_timeout(id_tasks, HTTP_POLLING_SECONDS); + + if (is_connection_closed()) { + cancel_tasks(id_tasks); + return; + } + + if (result == nullptr) { + continue; // retry + } if (result->is_error()) { error_handler(result->to_json()); @@ -2927,7 +3047,7 @@ struct server_context { // make sure we're in the right embedding mode llama_set_embeddings(ctx, slot_batched->is_non_causal()); // apply lora, only need to do it once per batch - common_lora_adapters_apply(ctx, slot_batched->lora); + common_set_adapter_lora(ctx, slot_batched->lora); } // process the created batch of tokens @@ -3129,12 +3249,12 @@ struct server_context { json model_meta() const { return json { - {"vocab_type", llama_vocab_type (model)}, - {"n_vocab", llama_n_vocab (model)}, - {"n_ctx_train", llama_n_ctx_train (model)}, - {"n_embd", llama_n_embd (model)}, - {"n_params", llama_model_n_params(model)}, - {"size", llama_model_size (model)}, + {"vocab_type", llama_vocab_type (vocab)}, + {"n_vocab", llama_vocab_n_tokens (vocab)}, + {"n_ctx_train", llama_model_n_ctx_train(model)}, + {"n_embd", llama_model_n_embd (model)}, + {"n_params", llama_model_n_params (model)}, + {"size", llama_model_size (model)}, }; } }; @@ -3601,9 +3721,12 @@ int main(int argc, char ** argv) { { "default_generation_settings", ctx_server.default_generation_settings_for_props }, { "total_slots", ctx_server.params_base.n_parallel }, { "model_path", ctx_server.params_base.model }, - { "chat_template", common_get_builtin_chat_template(ctx_server.model) }, + { "chat_template", ctx_server.chat_templates.template_default->source() }, { "build_info", build_info }, }; + if (ctx_server.params_base.use_jinja && ctx_server.chat_templates.template_tool_use) { + data["chat_template_tool_use"] = ctx_server.chat_templates.template_tool_use->source(); + } res_ok(res, data); }; @@ -3626,6 +3749,7 @@ int main(int argc, char ** argv) { const auto handle_completions_impl = [&ctx_server, &res_error, &res_ok]( server_task_type type, json & data, + std::function is_connection_closed, httplib::Response & res, oaicompat_type oaicompat) { GGML_ASSERT(type == SERVER_TASK_TYPE_COMPLETION || type == SERVER_TASK_TYPE_INFILL); @@ -3639,7 +3763,7 @@ int main(int argc, char ** argv) { std::vector tasks; try { - std::vector tokenized_prompts = tokenize_input_prompts(ctx_server.ctx, data.at("prompt"), true, true); + std::vector tokenized_prompts = tokenize_input_prompts(ctx_server.vocab, data.at("prompt"), true, true); tasks.reserve(tokenized_prompts.size()); for (size_t i = 0; i < tokenized_prompts.size(); i++) { server_task task = server_task(type); @@ -3649,7 +3773,6 @@ int main(int argc, char ** argv) { task.prompt_tokens = std::move(tokenized_prompts[i]); task.params = server_task::params_from_json_cmpl( - ctx_server.model, ctx_server.ctx, ctx_server.params_base, data); @@ -3688,7 +3811,7 @@ int main(int argc, char ** argv) { } }, [&](const json & error_data) { res_error(res, error_data); - }); + }, is_connection_closed); ctx_server.queue_results.remove_waiting_task_ids(task_ids); } else { @@ -3698,6 +3821,7 @@ int main(int argc, char ** argv) { if (res_json.is_array()) { for (const auto & res : res_json) { if (!server_sent_event(sink, "data", res)) { + // sending failed (HTTP connection closed), cancel the generation return false; } } @@ -3707,6 +3831,9 @@ int main(int argc, char ** argv) { } }, [&](const json & error_data) { server_sent_event(sink, "error", error_data); + }, [&sink]() { + // note: do not use req.is_connection_closed here because req is already destroyed + return !sink.is_writable(); }); if (oaicompat != OAICOMPAT_TYPE_NONE) { static const std::string ev_done = "data: [DONE]\n\n"; @@ -3729,6 +3856,7 @@ int main(int argc, char ** argv) { return handle_completions_impl( SERVER_TASK_TYPE_COMPLETION, data, + req.is_connection_closed, res, OAICOMPAT_TYPE_NONE); }; @@ -3738,6 +3866,7 @@ int main(int argc, char ** argv) { return handle_completions_impl( SERVER_TASK_TYPE_COMPLETION, data, + req.is_connection_closed, res, OAICOMPAT_TYPE_COMPLETION); }; @@ -3745,13 +3874,13 @@ int main(int argc, char ** argv) { const auto handle_infill = [&ctx_server, &res_error, &handle_completions_impl](const httplib::Request & req, httplib::Response & res) { // check model compatibility std::string err; - if (llama_token_fim_pre(ctx_server.model) == LLAMA_TOKEN_NULL) { + if (llama_vocab_fim_pre(ctx_server.vocab) == LLAMA_TOKEN_NULL) { err += "prefix token is missing. "; } - if (llama_token_fim_suf(ctx_server.model) == LLAMA_TOKEN_NULL) { + if (llama_vocab_fim_suf(ctx_server.vocab) == LLAMA_TOKEN_NULL) { err += "suffix token is missing. "; } - if (llama_token_fim_mid(ctx_server.model) == LLAMA_TOKEN_NULL) { + if (llama_vocab_fim_mid(ctx_server.vocab) == LLAMA_TOKEN_NULL) { err += "middle token is missing. "; } if (!err.empty()) { @@ -3797,10 +3926,10 @@ int main(int argc, char ** argv) { data["input_extra"] = input_extra; // default to empty array if it's not exist std::string prompt = json_value(data, "prompt", std::string()); - std::vector tokenized_prompts = tokenize_input_prompts(ctx_server.ctx, prompt, true, true); + std::vector tokenized_prompts = tokenize_input_prompts(ctx_server.vocab, prompt, false, true); SRV_DBG("creating infill tasks, n_prompts = %d\n", (int) tokenized_prompts.size()); data["prompt"] = format_infill( - ctx_server.ctx, + ctx_server.vocab, data.at("input_prefix"), data.at("input_suffix"), data.at("input_extra"), @@ -3814,6 +3943,7 @@ int main(int argc, char ** argv) { return handle_completions_impl( SERVER_TASK_TYPE_INFILL, data, + req.is_connection_closed, res, OAICOMPAT_TYPE_NONE); // infill is not OAI compatible }; @@ -3824,10 +3954,14 @@ int main(int argc, char ** argv) { return; } - json data = oaicompat_chat_completion_params_parse(ctx_server.model, json::parse(req.body), params.chat_template); + auto body = json::parse(req.body); + const auto & chat_template = body.contains("tools") && ctx_server.chat_templates.template_tool_use ? *ctx_server.chat_templates.template_tool_use : *ctx_server.chat_templates.template_default; + json data = oaicompat_completion_params_parse(body, chat_template, params.use_jinja); + return handle_completions_impl( SERVER_TASK_TYPE_COMPLETION, data, + req.is_connection_closed, res, OAICOMPAT_TYPE_CHAT); }; @@ -3857,7 +3991,7 @@ int main(int argc, char ** argv) { const bool add_special = json_value(body, "add_special", false); const bool with_pieces = json_value(body, "with_pieces", false); - llama_tokens tokens = tokenize_mixed(ctx_server.ctx, body.at("content"), add_special, true); + llama_tokens tokens = tokenize_mixed(ctx_server.vocab, body.at("content"), add_special, true); if (with_pieces) { for (const auto& token : tokens) { @@ -3933,7 +4067,7 @@ int main(int argc, char ** argv) { } } - std::vector tokenized_prompts = tokenize_input_prompts(ctx_server.ctx, prompt, true, true); + std::vector tokenized_prompts = tokenize_input_prompts(ctx_server.vocab, prompt, true, true); for (const auto & tokens : tokenized_prompts) { // this check is necessary for models that do not add BOS token to the input if (tokens.empty()) { @@ -3974,7 +4108,7 @@ int main(int argc, char ** argv) { }, [&](const json & error_data) { res_error(res, error_data); error = true; - }); + }, req.is_connection_closed); ctx_server.queue_results.remove_waiting_task_ids(task_ids); } @@ -4033,20 +4167,20 @@ int main(int argc, char ** argv) { return; } - llama_tokens tokenized_query = tokenize_input_prompts(ctx_server.ctx, query, /* add_special */ false, true)[0]; + llama_tokens tokenized_query = tokenize_input_prompts(ctx_server.vocab, query, /* add_special */ false, true)[0]; // create and queue the task json responses = json::array(); bool error = false; { std::vector tasks; - std::vector tokenized_docs = tokenize_input_prompts(ctx_server.ctx, documents, /* add_special */ false, true); + std::vector tokenized_docs = tokenize_input_prompts(ctx_server.vocab, documents, /* add_special */ false, true); tasks.reserve(tokenized_docs.size()); for (size_t i = 0; i < tokenized_docs.size(); i++) { server_task task = server_task(SERVER_TASK_TYPE_RERANK); task.id = ctx_server.queue_tasks.get_new_id(); task.index = i; - task.prompt_tokens = format_rerank(ctx_server.model, tokenized_query, tokenized_docs[i]); + task.prompt_tokens = format_rerank(ctx_server.vocab, tokenized_query, tokenized_docs[i]); tasks.push_back(task); } @@ -4064,7 +4198,7 @@ int main(int argc, char ** argv) { }, [&](const json & error_data) { res_error(res, error_data); error = true; - }); + }, req.is_connection_closed); } if (error) { @@ -4233,7 +4367,7 @@ int main(int argc, char ** argv) { // if a custom chat template is not supplied, we will use the one that comes with the model (if any) if (params.chat_template.empty()) { - if (!ctx_server.validate_builtin_chat_template()) { + if (!ctx_server.validate_builtin_chat_template(params.use_jinja)) { LOG_WRN("%s: The chat template that comes with this model is not yet supported, falling back to chatml. This may cause the model to output suboptimal responses\n", __func__); params.chat_template = "chatml"; } @@ -4241,8 +4375,8 @@ int main(int argc, char ** argv) { // print sample chat example to make it clear which template is used LOG_INF("%s: chat template, chat_template: %s, example_format: '%s'\n", __func__, - params.chat_template.empty() ? "(built-in)" : params.chat_template.c_str(), - common_chat_format_example(ctx_server.model, params.chat_template).c_str()); + ctx_server.chat_templates.template_default->source().c_str(), + common_chat_format_example(*ctx_server.chat_templates.template_default, ctx_server.params_base.use_jinja).c_str()); ctx_server.queue_tasks.on_new_task(std::bind( &server_context::process_single_task, &ctx_server, std::placeholders::_1)); diff --git a/examples/server/tests/unit/test_chat_completion.py b/examples/server/tests/unit/test_chat_completion.py index b15dba6eb..2e15348dc 100644 --- a/examples/server/tests/unit/test_chat_completion.py +++ b/examples/server/tests/unit/test_chat_completion.py @@ -4,22 +4,26 @@ from utils import * server = ServerPreset.tinyllama2() - -@pytest.fixture(scope="module", autouse=True) +@pytest.fixture(autouse=True) def create_server(): global server server = ServerPreset.tinyllama2() @pytest.mark.parametrize( - "model,system_prompt,user_prompt,max_tokens,re_content,n_prompt,n_predicted,finish_reason", + "model,system_prompt,user_prompt,max_tokens,re_content,n_prompt,n_predicted,finish_reason,jinja,chat_template", [ - (None, "Book", "What is the best book", 8, "(Suddenly)+", 77, 8, "length"), - ("codellama70b", "You are a coding assistant.", "Write the fibonacci function in c++.", 128, "(Aside|she|felter|alonger)+", 104, 64, "length"), + (None, "Book", "What is the best book", 8, "(Suddenly)+", 77, 8, "length", False, None), + (None, "Book", "What is the best book", 8, "(Suddenly)+", 77, 8, "length", True, None), + (None, "Book", "What is the best book", 8, "^ blue", 23, 8, "length", True, "This is not a chat template, it is"), + ("codellama70b", "You are a coding assistant.", "Write the fibonacci function in c++.", 128, "(Aside|she|felter|alonger)+", 104, 64, "length", False, None), + ("codellama70b", "You are a coding assistant.", "Write the fibonacci function in c++.", 128, "(Aside|she|felter|alonger)+", 104, 64, "length", True, None), ] ) -def test_chat_completion(model, system_prompt, user_prompt, max_tokens, re_content, n_prompt, n_predicted, finish_reason): +def test_chat_completion(model, system_prompt, user_prompt, max_tokens, re_content, n_prompt, n_predicted, finish_reason, jinja, chat_template): global server + server.jinja = jinja + server.chat_template = chat_template server.start() res = server.make_request("POST", "/chat/completions", data={ "model": model, diff --git a/examples/server/tests/unit/test_completion.py b/examples/server/tests/unit/test_completion.py index e5e3b6077..c1fc12462 100644 --- a/examples/server/tests/unit/test_completion.py +++ b/examples/server/tests/unit/test_completion.py @@ -1,4 +1,5 @@ import pytest +import requests import time from openai import OpenAI from utils import * @@ -405,3 +406,23 @@ def test_n_probs_post_sampling(): assert "bytes" in prob and type(prob["bytes"]) == list # because the test model usually output token with either 100% or 0% probability, we need to check all the top_probs assert any(prob["prob"] == 1.0 for prob in tok["top_probs"]) + + +def test_cancel_request(): + global server + server.n_ctx = 4096 + server.n_predict = -1 + server.n_slots = 1 + server.server_slots = True + server.start() + # send a request that will take a long time, but cancel it before it finishes + try: + server.make_request("POST", "/completion", data={ + "prompt": "I believe the meaning of life is", + }, timeout=0.1) + except requests.exceptions.ReadTimeout: + pass # expected + # make sure the slot is free + time.sleep(1) # wait for HTTP_POLLING_SECONDS + res = server.make_request("GET", "/slots") + assert res.body[0]["is_processing"] == False diff --git a/examples/server/tests/unit/test_infill.py b/examples/server/tests/unit/test_infill.py index ad4b8192a..10554db0f 100644 --- a/examples/server/tests/unit/test_infill.py +++ b/examples/server/tests/unit/test_infill.py @@ -18,7 +18,7 @@ def test_infill_without_input_extra(): "input_suffix": "}\n", }) assert res.status_code == 200 - assert match_regex("(Ann|small|shiny)+", res.body["content"]) + assert match_regex("(Ann|small|shiny|Daddy)+", res.body["content"]) def test_infill_with_input_extra(): diff --git a/examples/server/tests/utils.py b/examples/server/tests/utils.py index a1a94d0f1..9964db2f9 100644 --- a/examples/server/tests/utils.py +++ b/examples/server/tests/utils.py @@ -26,6 +26,9 @@ from re import RegexFlag import wget +DEFAULT_HTTP_TIMEOUT = 10 if "LLAMA_SANITIZE" not in os.environ else 30 + + class ServerResponse: headers: dict status_code: int @@ -69,13 +72,14 @@ class ServerProcess: pooling: str | None = None draft: int | None = None api_key: str | None = None - response_format: str | None = None lora_files: List[str] | None = None disable_ctx_shift: int | None = False draft_min: int | None = None draft_max: int | None = None no_webui: bool | None = None + jinja: bool | None = None chat_template: str | None = None + chat_template_file: str | None = None # session variables process: subprocess.Popen | None = None @@ -88,7 +92,7 @@ class ServerProcess: if "PORT" in os.environ: self.server_port = int(os.environ["PORT"]) - def start(self, timeout_seconds: int = 10) -> None: + def start(self, timeout_seconds: int | None = DEFAULT_HTTP_TIMEOUT) -> None: if "LLAMA_SERVER_BIN_PATH" in os.environ: server_path = os.environ["LLAMA_SERVER_BIN_PATH"] elif os.name == "nt": @@ -166,8 +170,12 @@ class ServerProcess: server_args.extend(["--draft-min", self.draft_min]) if self.no_webui: server_args.append("--no-webui") + if self.jinja: + server_args.append("--jinja") if self.chat_template: server_args.extend(["--chat-template", self.chat_template]) + if self.chat_template_file: + server_args.extend(["--chat-template-file", self.chat_template_file]) args = [str(arg) for arg in [server_path, *server_args]] print(f"bench: starting server with: {' '.join(args)}") @@ -219,17 +227,18 @@ class ServerProcess: path: str, data: dict | Any | None = None, headers: dict | None = None, + timeout: float | None = None, ) -> ServerResponse: url = f"http://{self.server_host}:{self.server_port}{path}" parse_body = False if method == "GET": - response = requests.get(url, headers=headers) + response = requests.get(url, headers=headers, timeout=timeout) parse_body = True elif method == "POST": - response = requests.post(url, headers=headers, json=data) + response = requests.post(url, headers=headers, json=data, timeout=timeout) parse_body = True elif method == "OPTIONS": - response = requests.options(url, headers=headers) + response = requests.options(url, headers=headers, timeout=timeout) else: raise ValueError(f"Unimplemented method: {method}") result = ServerResponse() diff --git a/examples/server/utils.hpp b/examples/server/utils.hpp index dc6e6e67e..c5987250c 100644 --- a/examples/server/utils.hpp +++ b/examples/server/utils.hpp @@ -16,6 +16,8 @@ // Change JSON_ASSERT from assert() to GGML_ASSERT: #define JSON_ASSERT GGML_ASSERT #include "json.hpp" +#include "minja.hpp" +#include "chat-template.hpp" #include #include @@ -118,7 +120,7 @@ static json json_get_nested_values(const std::vector & paths, const * - only string, example: "string" * - mixed string and tokens, example: [12, 34, "string", 56, 78] */ -static llama_tokens tokenize_mixed(const llama_context * ctx, const json & json_prompt, bool add_special, bool parse_special) { +static llama_tokens tokenize_mixed(const llama_vocab * vocab, const json & json_prompt, bool add_special, bool parse_special) { // If `add_bos` is true, we only add BOS, when json_prompt is a string, // or the first element of the json_prompt array is a string. llama_tokens prompt_tokens; @@ -131,10 +133,10 @@ static llama_tokens tokenize_mixed(const llama_context * ctx, const json & json_ llama_tokens p; if (first) { - p = common_tokenize(ctx, s, add_special, parse_special); + p = common_tokenize(vocab, s, add_special, parse_special); first = false; } else { - p = common_tokenize(ctx, s, false, parse_special); + p = common_tokenize(vocab, s, false, parse_special); } prompt_tokens.insert(prompt_tokens.end(), p.begin(), p.end()); @@ -148,7 +150,7 @@ static llama_tokens tokenize_mixed(const llama_context * ctx, const json & json_ } } else { auto s = json_prompt.template get(); - prompt_tokens = common_tokenize(ctx, s, add_special, parse_special); + prompt_tokens = common_tokenize(vocab, s, add_special, parse_special); } return prompt_tokens; @@ -166,11 +168,11 @@ static llama_tokens tokenize_mixed(const llama_context * ctx, const json & json_ * - "prompt": [[12, 34, 56], [78, 90, 12]] * - "prompt": [[12, 34, "string", 56, 78], [12, 34, 56]] */ -static std::vector tokenize_input_prompts(llama_context * ctx, const json & json_prompt, bool add_special, bool parse_special) { +static std::vector tokenize_input_prompts(const llama_vocab * vocab, const json & json_prompt, bool add_special, bool parse_special) { std::vector result; if (json_prompt.is_string() || json_is_array_of_mixed_numbers_strings(json_prompt)) { // string or mixed - result.push_back(tokenize_mixed(ctx, json_prompt, add_special, parse_special)); + result.push_back(tokenize_mixed(vocab, json_prompt, add_special, parse_special)); } else if (json_is_array_of_numbers(json_prompt)) { // array of tokens result.push_back(json_prompt.get()); @@ -179,7 +181,7 @@ static std::vector tokenize_input_prompts(llama_context * ctx, con result.reserve(json_prompt.size()); for (const auto & p : json_prompt) { if (p.is_string() || json_is_array_of_mixed_numbers_strings(p)) { - result.push_back(tokenize_mixed(ctx, p, add_special, parse_special)); + result.push_back(tokenize_mixed(vocab, p, add_special, parse_special)); } else if (json_is_array_of_numbers(p)) { // array of tokens result.push_back(p.get()); @@ -231,21 +233,23 @@ static size_t validate_utf8(const std::string& text) { // // format rerank task: [BOS]query[EOS][SEP]doc[EOS] -static llama_tokens format_rerank(const struct llama_model * model, const llama_tokens & query, const llama_tokens & doc) { +static llama_tokens format_rerank(const struct llama_vocab * vocab, const llama_tokens & query, const llama_tokens & doc) { llama_tokens result; + result.reserve(doc.size() + query.size() + 4); - result.push_back(llama_token_bos(model)); + result.push_back(llama_vocab_bos(vocab)); result.insert(result.end(), query.begin(), query.end()); - result.push_back(llama_token_eos(model)); - result.push_back(llama_token_sep(model)); + result.push_back(llama_vocab_eos(vocab)); + result.push_back(llama_vocab_sep(vocab)); result.insert(result.end(), doc.begin(), doc.end()); - result.push_back(llama_token_eos(model)); + result.push_back(llama_vocab_eos(vocab)); + return result; } // format infill task static llama_tokens format_infill( - const llama_context * ctx, + const llama_vocab * vocab, const json & input_prefix, const json & input_suffix, const json & input_extra, @@ -272,15 +276,14 @@ static llama_tokens format_infill( llama_tokens extra_tokens; extra_tokens.reserve(n_ctx); - auto model = llama_get_model(ctx); - auto tokens_prefix = tokenize_mixed(ctx, input_prefix, false, false); - auto tokens_suffix = tokenize_mixed(ctx, input_suffix, false, false); + auto tokens_prefix = tokenize_mixed(vocab, input_prefix, false, false); + auto tokens_suffix = tokenize_mixed(vocab, input_suffix, false, false); - if (llama_token_fim_rep(model) != LLAMA_TOKEN_NULL) { + if (llama_vocab_fim_rep(vocab) != LLAMA_TOKEN_NULL) { // TODO: make project name an input - static const auto k_fim_repo = common_tokenize(ctx, "myproject\n", false, false); + static const auto k_fim_repo = common_tokenize(vocab, "myproject\n", false, false); - extra_tokens.push_back(llama_token_fim_rep(model)); + extra_tokens.push_back(llama_vocab_fim_rep(vocab)); extra_tokens.insert(extra_tokens.end(), k_fim_repo.begin(), k_fim_repo.end()); } for (const auto & chunk : input_extra) { @@ -288,28 +291,28 @@ static llama_tokens format_infill( const std::string text = json_value(chunk, "text", std::string()); const std::string filename = json_value(chunk, "filename", std::string("tmp")); - if (llama_token_fim_sep(model) != LLAMA_TOKEN_NULL) { - const auto k_fim_file = common_tokenize(ctx, filename + "\n", false, false); + if (llama_vocab_fim_sep(vocab) != LLAMA_TOKEN_NULL) { + const auto k_fim_file = common_tokenize(vocab, filename + "\n", false, false); - extra_tokens.insert(extra_tokens.end(), llama_token_fim_sep(model)); + extra_tokens.insert(extra_tokens.end(), llama_vocab_fim_sep(vocab)); extra_tokens.insert(extra_tokens.end(), k_fim_file.begin(), k_fim_file.end()); } else { // chunk separator in binary form to avoid confusing the AI static const char k_chunk_prefix_str[] = {0x0a, 0x0a, 0x2d, 0x2d, 0x2d, 0x20, 0x73, 0x6e, 0x69, 0x70, 0x70, 0x65, 0x74, 0x20, 0x2d, 0x2d, 0x2d, 0x0a, 0x0a, 0x00}; - static const auto k_chunk_prefix_tokens = common_tokenize(ctx, k_chunk_prefix_str, false, false); + static const auto k_chunk_prefix_tokens = common_tokenize(vocab, k_chunk_prefix_str, false, false); extra_tokens.insert(extra_tokens.end(), k_chunk_prefix_tokens.begin(), k_chunk_prefix_tokens.end()); } - const auto chunk_tokens = common_tokenize(ctx, text, false, false); + const auto chunk_tokens = common_tokenize(vocab, text, false, false); extra_tokens.insert(extra_tokens.end(), chunk_tokens.begin(), chunk_tokens.end()); } - if (llama_token_fim_sep(model) != LLAMA_TOKEN_NULL) { + if (llama_vocab_fim_sep(vocab) != LLAMA_TOKEN_NULL) { // TODO: current filename - static const auto k_fim_file = common_tokenize(ctx, "filename\n", false, false); + static const auto k_fim_file = common_tokenize(vocab, "filename\n", false, false); - extra_tokens.insert(extra_tokens.end(), llama_token_fim_sep(model)); + extra_tokens.insert(extra_tokens.end(), llama_vocab_fim_sep(vocab)); extra_tokens.insert(extra_tokens.end(), k_fim_file.begin(), k_fim_file.end()); } @@ -325,15 +328,15 @@ static llama_tokens format_infill( tokens_prefix.erase(tokens_prefix.begin(), tokens_prefix.begin() + tokens_prefix.size() - n_prefix_take); tokens_suffix.resize(n_suffix_take); - tokens_prefix.insert(tokens_prefix.begin(), llama_token_fim_pre(model)); + tokens_prefix.insert(tokens_prefix.begin(), llama_vocab_fim_pre(vocab)); tokens_prefix.insert(tokens_prefix.end(), tokens_prompt.begin(), tokens_prompt.end()); - tokens_suffix.insert(tokens_suffix.begin(), llama_token_fim_suf(model)); + tokens_suffix.insert(tokens_suffix.begin(), llama_vocab_fim_suf(vocab)); auto embd_inp = spm_infill ? tokens_suffix : tokens_prefix; auto embd_end = spm_infill ? tokens_prefix : tokens_suffix; - if (llama_add_bos_token(model)) { - embd_inp.insert(embd_inp.begin(), llama_token_bos(model)); + if (llama_vocab_get_add_bos(vocab)) { + embd_inp.insert(embd_inp.begin(), llama_vocab_bos(vocab)); } SRV_DBG("extra: n_ctx = %d, n_extra_take = %d, n_extra = %d\n", n_ctx, n_extra_take, (int) extra_tokens.size()); @@ -342,13 +345,13 @@ static llama_tokens format_infill( embd_inp.insert(embd_inp.begin(), extra_tokens.end() - n_extra_take, extra_tokens.end()); embd_inp.insert(embd_inp.end(), embd_end.begin(), embd_end.end()); - embd_inp.push_back(llama_token_fim_mid(model)); + embd_inp.push_back(llama_vocab_fim_mid(vocab)); return embd_inp; } // Format given chat. If tmpl is empty, we take the template from model metadata -inline std::string format_chat(const struct llama_model * model, const std::string & tmpl, const std::vector & messages) { +inline std::string format_chat(const common_chat_template & tmpl, const std::vector & messages) { std::vector chat; for (size_t i = 0; i < messages.size(); ++i) { @@ -376,7 +379,7 @@ inline std::string format_chat(const struct llama_model * model, const std::stri chat.push_back({role, content}); } - const auto formatted_chat = common_chat_apply_template(model, tmpl, chat, true); + const auto formatted_chat = common_chat_apply_template(tmpl, chat, true, /* use_jinja= */ false); LOG_DBG("formatted_chat: '%s'\n", formatted_chat.c_str()); return formatted_chat; @@ -507,7 +510,7 @@ static std::string tokens_to_str(llama_context * ctx, Iter begin, Iter end) { // format incomplete utf-8 multibyte character for output static std::string tokens_to_output_formatted_string(const llama_context * ctx, const llama_token token) { - std::string out = token == -1 ? "" : common_token_to_piece(ctx, token); + std::string out = token == LLAMA_TOKEN_NULL ? "" : common_token_to_piece(ctx, token); // if the size is 1 and first bit is 1, meaning it's a partial character // (size > 1 meaning it's already a known token) @@ -575,14 +578,23 @@ static json oaicompat_completion_params_parse(const json & body) { return llama_params; } -static json oaicompat_chat_completion_params_parse( - const struct llama_model * model, - const json & body, /* openai api json semantics */ - const std::string & chat_template) { +static json oaicompat_completion_params_parse( + const json & body, /* openai api json semantics */ + const common_chat_template & tmpl, + bool use_jinja) +{ json llama_params; - // Apply chat template to the list of messages - llama_params["prompt"] = format_chat(model, chat_template, body.at("messages")); + auto tools = json_value(body, "tools", json()); + auto has_tools = tools.is_array() && !tools.empty(); + + if (has_tools) { + if (use_jinja) { + LOG_WRN("tools param is not fully supported yet\n"); + } else { + throw std::runtime_error("tools param requires --jinja flag"); + } + } // Handle "stop" field if (body.contains("stop") && body.at("stop").is_string()) { @@ -605,6 +617,13 @@ static json oaicompat_chat_completion_params_parse( } } + // Apply chat template to the list of messages + if (use_jinja) { + llama_params["prompt"] = tmpl.apply(body.at("messages"), tools, /* add_generation_prompt= */ true); + } else { + llama_params["prompt"] = format_chat(tmpl, body.at("messages")); + } + // Handle "n" field int n_choices = json_value(body, "n", 1); if (n_choices != 1) { @@ -620,7 +639,7 @@ static json oaicompat_chat_completion_params_parse( } // Params supported by OAI but unsupported by llama.cpp - static const std::vector unsupported_params { "tools", "tool_choice" }; + static const std::vector unsupported_params { "tool_choice" }; for (const auto & param : unsupported_params) { if (body.contains(param)) { throw std::runtime_error("Unsupported param: " + param); @@ -764,14 +783,18 @@ static json format_logit_bias(const std::vector & logit_bias) return data; } -static std::string safe_json_to_str(json data) { +static std::string safe_json_to_str(const json & data) { return data.dump(-1, ' ', false, json::error_handler_t::replace); } static std::vector get_token_probabilities(llama_context * ctx, int idx) { std::vector cur; const auto * logits = llama_get_logits_ith(ctx, idx); - const int n_vocab = llama_n_vocab(llama_get_model(ctx)); + + const llama_model * model = llama_get_model(ctx); + const llama_vocab * vocab = llama_model_get_vocab(model); + + const int n_vocab = llama_vocab_n_tokens(vocab); cur.resize(n_vocab); for (llama_token token_id = 0; token_id < n_vocab; token_id++) { @@ -799,8 +822,8 @@ static std::vector get_token_probabilities(llama_context * ctx } static bool are_lora_equal( - const std::vector & l1, - const std::vector & l2) { + const std::vector & l1, + const std::vector & l2) { if (l1.size() != l2.size()) { return false; } @@ -814,10 +837,10 @@ static bool are_lora_equal( } // parse lora config from JSON request, returned a copy of lora_base with updated scale -static std::vector parse_lora_request( - const std::vector & lora_base, +static std::vector parse_lora_request( + const std::vector & lora_base, const json & data) { - std::vector lora(lora_base); + std::vector lora(lora_base); int max_idx = lora.size(); // clear existing value diff --git a/examples/server/webui/index.html b/examples/server/webui/index.html index dcdd41079..d3893ea4e 100644 --- a/examples/server/webui/index.html +++ b/examples/server/webui/index.html @@ -37,7 +37,7 @@
+ }" @click="setViewingConv(conv.id)" dir="auto"> {{ conv.messages[0].content }}
@@ -62,53 +62,57 @@
- + - +
- +
+ +
-
-
- - - +
+
+
+ + + +
+
    +
  • + +
    • +
  • + +
    • +
-
    -
  • - -
    • -
  • - -
    • -
@@ -137,6 +141,7 @@ :msg="pendingMsg" :key="pendingMsg.id" :is-generating="isGenerating" + :show-thought-in-progress="config.showThoughtInProgress" :edit-user-msg-and-regenerate="() => {}" :regenerate-msg="() => {}">
@@ -152,6 +157,7 @@ @keydown.enter.shift.exact.prevent="inputMsg += '\n'" :disabled="isGenerating" id="msg-input" + dir="auto" > @@ -197,6 +203,20 @@
+ +
+ Reasoning models +
+
+ + Expand though process by default for generating message +
+
+ + Exclude thought process when sending request to API (Recommended for DeepSeek-R1) +
+
+
Advanced config @@ -244,6 +264,7 @@