Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2025-01-27 20:43:07 +01:00
Code Issues Packages Projects Releases Wiki Activity

Merge remote-tracking branch 'upstream/master' into Remove_obsolete_HIP_workaround

Browse Source
This commit is contained in:
Nikita Sarychev 2025-01-25 16:46:43 -08:00
commit 7088822e5a
240 changed files with 22244 additions and 10380 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
.devops/cpu.Dockerfile
View File

@ -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 && \

12
.github/ISSUE_TEMPLATE/010-bug-compilation.yml vendored
View File

@ -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:

12
.github/ISSUE_TEMPLATE/019-bug-misc.yml vendored
View File

@ -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:

160
.github/workflows/build.yml vendored
View File

@ -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)

3
.github/workflows/docker.yml vendored
View File

@ -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

4
.github/workflows/editorconfig.yml vendored
View File

@ -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

25
.github/workflows/server.yml vendored
View File

@ -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' }}

1
.gitignore vendored
View File

@ -18,6 +18,7 @@
*.metallib
*.o
*.so
*.swp
*.tmp
# IDE / OS

75
CMakeLists.txt
View File

@ -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 ${CMAKE_BINARY_DIR}/bin)
if (CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR)
set(LLAMA_STANDALONE ON)
@ -49,6 +50,7 @@ endif()
if (MSVC)
add_compile_options("$<$<COMPILE_LANGUAGE:C>:/utf-8>")
add_compile_options("$<$<COMPILE_LANGUAGE:CXX>:/utf-8>")
add_compile_options(/bigobj)
endif()
#
@ -83,11 +85,8 @@ include(${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()

8
CODEOWNERS
View File

@ -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

102
CONTRIBUTING.md
View File

@ -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 `<class>_<method>`, with `<method>` being `<action>_<noun>`
```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` `<action>` can be omitted
- The `<noun>` can be omitted if not necessary
- The `_context` suffix of the `<class>` is optional. Use it to disambiguate symbols when needed
- Use `init`/`free` for constructor/destructor `<action>`
- 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:

2
Makefile
View File

@ -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)

45
README.md
View File

@ -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
</details>
@ -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 <user>/<model>[: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.
- <details open>
<summary>Run simple text completion</summary>
```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.
```
</details>
- <details>
<summary>Run in conversation mode</summary>
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
</details>
- <details>
<summary>Run with custom chat template</summary>
<summary>Run in conversation mode with custom chat template</summary>
```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)
</details>
- <details>
<summary>Run simple text completion</summary>
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.
```
</details>

66
ci/run.sh
View File

@ -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

2
cmake/build-info.cmake
View File

@ -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
)

2
common/CMakeLists.txt
View File

@ -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

157
common/arg.cpp
View File

@ -22,6 +22,11 @@ common_arg & common_arg::set_examples(std::initializer_list<enum llama_example>
return *this;
}
common_arg & common_arg::set_excludes(std::initializer_list<enum llama_example> 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<std::string>(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<ggml_backend_dev_t> parse_device_list(const std::string & val
return devices;
}
static void add_rpc_devices(std::string servers) {
auto rpc_servers = string_split<std::string>(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"}, "<user>/<model>[: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"}, "<user>/<model>[: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"}, "<user>/<model>[: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<char>(file),
std::istreambuf_iterator<char>(),
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(

3
common/arg.h
View File

@ -12,6 +12,7 @@
struct common_arg {
std::set<enum llama_example> examples = {LLAMA_EXAMPLE_COMMON};
std::set<enum llama_example> excludes = {};
std::vector<const char *> 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<enum llama_example> examples);
common_arg & set_excludes(std::initializer_list<enum llama_example> 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();

268
common/chat-template.hpp Normal file
View File

@ -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 <json.hpp>
#include <string>
#include <vector>
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<minja::TemplateNode> 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", "<System Needle>"}},
{{"role", "user"}, {"content", "Hey"}}
}, {}, false).find("<System Needle>") != 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

392
common/common.cpp
View File

@ -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 <algorithm>
#include <cinttypes>
@ -70,6 +74,22 @@
#include <sys/syslimits.h>
#endif
#define LLAMA_CURL_MAX_URL_LENGTH 2084 // Maximum URL Length in Chrome: 2083
//
// CURL utils
//
using curl_ptr = std::unique_ptr<CURL, decltype(&curl_easy_cleanup)>;
// 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<std::string> & 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<std::string> string_split(const std::string & str, const std::string & delimiter) {
std::vector<std::string> 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<llama_token> 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<common_lora_adapter_info> & lora) {
llama_lora_adapter_clear(ctx);
void common_set_adapter_lora(struct llama_context * ctx, std::vector<common_adapter_lora_info> & 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, decltype(&curl_easy_cleanup)> 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 <repo, file> (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<std::string, std::string> common_get_hf_file(const std::string & hf_repo_with_tag, const std::string & hf_token) {
auto parts = string_split<std::string>(hf_repo_with_tag, ':');
std::string tag = parts.size() > 1 ? parts.back() : "latest";
std::string hf_repo = parts[0];
if (string_split<std::string>(hf_repo, '/').size() != 2) {
throw std::invalid_argument("error: invalid HF repo format, expected <user>/<model>[: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<std::string *>(data)->append((char * ) ptr, size * nmemb);
return size * nmemb;
};
curl_easy_setopt(curl.get(), CURLOPT_WRITEFUNCTION, static_cast<CURLOPT_WRITEFUNCTION_PTR>(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<std::string, std::string> 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<llama_token> 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<llama_token> 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<llama_token> 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<llama_token> 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<llama_token> & tokens, bool special) {
std::string common_detokenize(const struct llama_context * ctx, const std::vector<llama_token> & 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<llama_token> & 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<llama_token
// Chat template utils
//
std::string common_get_builtin_chat_template(const struct llama_model * model) {
static const char * template_key = "tokenizer.chat_template";
// call with NULL buffer to get the total size of the string
int32_t res = llama_model_meta_val_str(model, template_key, NULL, 0);
if (res > 0) {
std::vector<char> 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, "<s>", "</s>");
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<common_chat_msg> & 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<llama_chat_message> 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<char> 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<common_chat_msg> & 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<common_chat_msg> 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<common_chat_msg> 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<minja::chat_template>(default_template_src, token_bos, token_eos),
template_tool_use_src.empty()
? nullptr
: std::make_unique<minja::chat_template>(template_tool_use_src, token_bos, token_eos)
};
}
//

96
common/common.h
View File

@ -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<llama_token>;
@ -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<std::string> in_files; // all input files
std::vector<std::string> antiprompt; // strings upon which more user input is prompted (a.k.a. reverse prompts)
std::vector<llama_model_kv_override> 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<common_lora_adapter_info> 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<common_adapter_lora_info> lora_adapters; // lora adapter path with user defined scale
std::vector<common_control_vector_load_info> 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<std::string> 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<std::string> 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<std::string> & values, const std::string & separator);
std::vector<std::string> 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<class T>
@ -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<llama_model_kv_override> & 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<llama_lora_adapter_ptr> lora;
std::vector<llama_adapter_lora_ptr> 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<std::string, std::string> 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<common_lora_adapter_info> & lora);
void common_set_adapter_lora(struct llama_context * ctx, std::vector<common_adapter_lora_info> & lora);
//
// Batch utils
@ -541,7 +568,7 @@ std::vector<llama_token> common_tokenize(
bool parse_special = false);
std::vector<llama_token> 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<llama_token> & tokens,
bool special = true);
std::string common_detokenize(
const struct llama_vocab * vocab,
const std::vector<llama_token> & 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<common_chat_template> template_default; // always set (defaults to chatml)
std::unique_ptr<common_chat_template> 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<common_chat_msg> & 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<common_chat_msg> & 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

98
common/json-schema-to-grammar.cpp
View File

@ -1,4 +1,6 @@
#include "json-schema-to-grammar.h"
#include "common.h"
#include <algorithm>
#include <fstream>
#include <map>
@ -11,11 +13,6 @@
using json = nlohmann::ordered_json;
template <typename Iterator>
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<int>::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<char, std::string> GRAMMAR_LITERAL_ESCAPES = {
std::unordered_set<char> NON_LITERAL_SET = {'|', '.', '(', ')', '[', ']', '{', '}', '*', '+', '?'};
std::unordered_set<char> ESCAPED_IN_REGEXPS_BUT_NOT_IN_LITERALS = {'^', '$', '.', '[', ']', '(', ')', '|', '{', '}', '*', '+', '?'};
template <typename Iterator>
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<std::string> split(const std::string & str, const std::string & delimiter) {
std::vector<std::string> 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<std::string(const std::smatch &)> & 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<void(const llama_grammar_builder &)> & cb);
std::function<json(const std::string &)> _fetch_json;
bool _dotall;
std::map<std::string, std::string> _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<int>::max();
try {
@ -854,7 +809,7 @@ public:
return;
}
std::string pointer = ref.substr(ref.find('#') + 1);
std::vector<std::string> tokens = split(pointer, "/");
std::vector<std::string> 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<void(const llama_grammar_builder &)> & 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();
}

10
common/json-schema-to-grammar.h
View File

@ -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<std::string(const std::string &, const std::string &)> add_rule;
std::function<std::string(const std::string &, const nlohmann::ordered_json &)> add_schema;
std::function<void(nlohmann::ordered_json &)> resolve_refs;
};
std::string build_grammar(const std::function<void(const llama_grammar_builder &)> & cb);

2812
common/minja.hpp Normal file
View File

File diff suppressed because it is too large Load Diff

24
common/ngram-cache.cpp
View File

@ -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<llama_token, int> 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<common_ngram> & 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<llama_token, int> 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;
}

4
common/ngram-cache.h
View File

@ -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;
}
}

17
common/sampling.cpp
View File

@ -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<llama_token>(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));

33
common/speculative.cpp
View File

@ -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());

221
convert_hf_to_gguf.py
View File

@ -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:

87
convert_hf_to_gguf_update.py
View File

@ -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"},
]

34
convert_lora_to_gguf.py
View File

@ -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)

2
docs/backend/SYCL.md
View File

@ -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*:

4
docs/build.md
View File

@ -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

317
docs/cuda-fedora.md Normal file
View File

@ -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)
---

10
docs/development/HOWTO-add-model.md
View File

@ -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/).

12
docs/docker.md
View File

@ -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.

6
examples/batched-bench/batched-bench.cpp
View File

@ -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();

6
examples/batched.swift/Sources/main.swift
View File

@ -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 {

16
examples/batched/batched.cpp
View File

@ -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<llama_token> 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();

8
examples/convert-llama2c-to-ggml/convert-llama2c-to-ggml.cpp
View File

@ -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;

12
examples/cvector-generator/cvector-generator.cpp
View File

@ -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];

8
examples/embedding/embedding.cpp
View File

@ -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<float> embeddings(n_embd_count * n_embd, 0);
float * emb = embeddings.data();

5
examples/eval-callback/eval-callback.cpp
View File

@ -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<llama_token> tokens = common_tokenize(ctx, params.prompt, add_bos);

23
examples/export-lora/export-lora.cpp
View File

@ -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 <map>
#include <vector>
#include <string>
#include <thread>
#include <fstream>
static bool g_verbose = false;
@ -128,7 +129,7 @@ struct lora_merge_ctx {
lora_merge_ctx(
std::string & base_fname,
std::vector<common_lora_adapter_info> & lora_files,
std::vector<common_adapter_lora_info> & 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];

1
examples/gguf-hash/gguf-hash.cpp
View File

@ -1,4 +1,5 @@
#include "ggml.h"
#include "gguf.h"
#include <cstdlib> /* abort() */
#include <cstddef>

14
examples/gguf-split/gguf-split.cpp
View File

@ -1,16 +1,18 @@
#include "ggml.h"
#include "gguf.h"
#include "llama.h"
#include "common.h"
#include <algorithm>
#include <cinttypes>
#include <climits>
#include <cstdio>
#include <cstdlib>
#include <stdexcept>
#include <cstring>
#include <fstream>
#include <string>
#include <vector>
#include <climits>
#include <cstdio>
#include <cstring>
#include <stdexcept>
#if defined(_WIN32)
#include <windows.h>
@ -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++;
}
}

10
examples/gguf-split/tests.sh
View File

@ -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

16
examples/gguf/gguf.cpp
View File

@ -1,10 +1,9 @@
#include "ggml.h"
#include "gguf.h"
#include <cstdio>
#include <cinttypes>
#include <string>
#include <sstream>
#include <fstream>
#include <vector>
#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]);

23
examples/gritlm/gritlm.cpp
View File

@ -11,6 +11,7 @@ static std::vector<std::vector<float>> encode(llama_context * ctx, const std::ve
std::vector<std::vector<float>> 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<std::vector<float>> encode(llama_context * ctx, const std::ve
const std::string input_string = instruction + sentences[i];
std::vector<llama_token> inputs = common_tokenize(model, input_string, true, false);
std::vector<llama_token> 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<std::vector<float>> 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<float> 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<llama_token> inputs = common_tokenize(model, prompt, false, true);
std::vector<llama_token> 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<std::vector<float>> d_rep = encode(ctx, documents, gritlm_instruction(""));
const std::vector<std::vector<float>> 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;

16
examples/imatrix/imatrix.cpp
View File

@ -7,7 +7,6 @@
#include <cstdio>
#include <cstring>
#include <ctime>
#include <sstream>
#include <thread>
#include <mutex>
#include <vector>
@ -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<std::string, Stats> 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);

40
examples/infill/infill.cpp
View File

@ -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<llama_token> embd_inp;
std::vector<llama_token> embd_end;
std::vector<llama_token> inp_pfx = common_tokenize(ctx, params.input_prefix, false);
std::vector<llama_token> 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<llama_token> inp_pfx = common_tokenize(ctx, params.input_prefix, false);
std::vector<llama_token> 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");

57
examples/llama-bench/llama-bench.cpp
View File

@ -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<std::string>(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<ggml_backend_dev_t> 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<llama_token> 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();

13
examples/llama.android/llama/src/main/cpp/llama-android.cpp
View File

@ -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<llama_model *>(model));
llama_model_free(reinterpret_cast<llama_model *>(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<llama_context *>(context_pointer);
const auto batch = reinterpret_cast<llama_batch *>(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<llama_batch *>(batch_pointer);
const auto sampler = reinterpret_cast<llama_sampler *>(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;
}

5
examples/llama.android/llama/src/main/java/android/llama/cpp/LLamaAndroid.kt
View File

@ -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<String> = flow {
fun send(message: String, formatChat: Boolean = false): Flow<String> = 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) {

6
examples/llama.swiftui/llama.cpp.swift/LibLlama.swift
View File

@ -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])

46
examples/llava/README-minicpmo2.6.md Normal file
View File

@ -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
```

35
examples/llava/clip.cpp
View File

@ -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<std::vector<clip_image_u8 *>> 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];

15
examples/llava/llava-cli.cpp
View File

@ -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 = "</s>";
} 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;
}

17
examples/llava/llava.cpp
View File

@ -216,7 +216,7 @@ static bool clip_llava_handle_patches(clip_ctx * ctx_clip, std::vector<float *>
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;

22
examples/llava/minicpmv-cli.cpp
View File

@ -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+"<image>").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 = "</s>";
} 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, "</s>") == 0) break;
if (strstr(tmp, "###")) break; // Yi-VL behavior
printf("%s", tmp);// mistral llava-1.6
if (strstr(response.c_str(), "<user>")) break; // minicpm-v
fflush(stdout);

15
examples/llava/minicpmv-convert-image-encoder-to-gguf.py
View File

@ -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:

2
examples/llava/minicpmv-surgery.py
View File

@ -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

19
examples/llava/qwen2vl-cli.cpp
View File

@ -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 = "</s>";
} 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;
}

6
examples/lookahead/lookahead.cpp
View File

@ -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<llama_token> inp;
std::vector<llama_token> 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;
}

4
examples/lookup/lookup.cpp
View File

@ -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<llama_token> 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;
}

4
examples/main/README.md
View File

@ -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

93
examples/main/main.cpp
View File

@ -4,8 +4,8 @@
#include "log.h"
#include "sampling.h"
#include "llama.h"
#include "chat-template.hpp"
#include <cassert>
#include <cstdio>
#include <cstring>
#include <ctime>
@ -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<common_chat_msg> & 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 = &params;
@ -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<llama_token> 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;
}

4
examples/parallel/parallel.cpp
View File

@ -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)) {

12
examples/passkey/passkey.cpp
View File

@ -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();

52
examples/perplexity/perplexity.cpp
View File

@ -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<winogrande_entry> 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<llama_token> 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<uint16_t> log_probs_uint16(size_t(n_ctx - 1 - n_ctx/2) * nv);
std::vector<float> 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",

10
examples/quantize-stats/quantize-stats.cpp
View File

@ -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();

4
examples/quantize/tests.sh
View File

@ -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

10
examples/retrieval/retrieval.cpp
View File

@ -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<float> embeddings(n_chunks * n_embd, 0);
float * emb = embeddings.data();

2
examples/run/CMakeLists.txt
View File

@ -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)

5
examples/run/README.md
View File

@ -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 <value>
Context size (default: 2048)
-n, --ngl <value>
-n, -ngl, --ngl <value>
Number of GPU layers (default: 0)
--temp <value>
Temperature (default: 0.8)

26
examples/run/linenoise.cpp/LICENSE Normal file
View File

@ -0,0 +1,26 @@
Copyright (c) 2010-2014, Salvatore Sanfilippo <antirez at gmail dot com>
Copyright (c) 2010-2013, Pieter Noordhuis <pcnoordhuis at gmail dot com>
Copyright (c) 2025, Eric Curtin <ericcurtin17 at gmail dot com>
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.

1350
examples/run/linenoise.cpp/linenoise.cpp Normal file
View File

File diff suppressed because it is too large Load Diff

128
examples/run/linenoise.cpp/linenoise.h Normal file
View File

@ -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 <antirez at gmail dot com>
* Copyright (c) 2010-2013, Pieter Noordhuis <pcnoordhuis at gmail dot com>
* Copyright (c) 2025, Eric Curtin <ericcurtin17 at gmail dot com>
*
* 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 <stddef.h> /* For size_t. */
#include <stdlib.h>
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 */

186
examples/run/run.cpp
View File

@ -11,19 +11,31 @@
# include <curl/curl.h>
#endif
#include <signal.h>
#include <climits>
#include <cstdarg>
#include <cstdio>
#include <cstring>
#include <filesystem>
#include <iostream>
#include <list>
#include <sstream>
#include <string>
#include <vector>
#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 <value>\n"
" Context size (default: %d)\n"
" -n, --ngl <value>\n"
" -n, -ngl, --ngl <value>\n"
" Number of GPU layers (default: %d)\n"
" --temp <value>\n"
" Temperature (default: %.1f)\n"
@ -526,7 +543,7 @@ class LlamaData {
llama_sampler_ptr sampler;
llama_context_ptr context;
std::vector<llama_chat_message> messages;
std::vector<std::string> msg_strs;
std::list<std::string> msg_strs;
std::vector<char> 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<std::string> 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<int>(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<llama_token> & 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<llama_token> & 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<llama_token> 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<char, decltype(&std::free)> line(const_cast<char *>(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<PHANDLER_ROUTINE>(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;
}

4
examples/save-load-state/save-load-state.cpp
View File

@ -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);

5
examples/server/README.md
View File

@ -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)<br/>(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)<br/>(env: LLAMA_ARG_FLASH_ATTN) |
| `-p, --prompt PROMPT` | prompt to start generation with |
| `--no-perf` | disable internal libllama performance timings (default: false)<br/>(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<br/>(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<br/>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**

1726
examples/server/httplib.h
View File

File diff suppressed because it is too large Load Diff

BIN
examples/server/public/index.html.gz
View File

Binary file not shown.

236
examples/server/server.cpp
View File

@ -19,6 +19,7 @@
#include "loading.html.hpp"
#include <atomic>
#include <chrono>
#include <condition_variable>
#include <cstddef>
#include <cinttypes>
@ -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<common_lora_adapter_info> lora;
std::vector<common_adapter_lora_info> lora;
std::vector<std::string> antiprompt;
std::vector<std::string> response_fields;
@ -198,15 +201,17 @@ struct server_task {
bool metrics_reset_bucket = false;
// used by SERVER_TASK_TYPE_SET_LORA
std::vector<common_lora_adapter_info> set_lora;
std::vector<common_adapter_lora_info> 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<std::string>(), false);
auto toks = common_tokenize(vocab, el[0].get<std::string>(), 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<common_lora_adapter_info> lora;
std::vector<common_adapter_lora_info> 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<std::mutex> 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<std::mutex> 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<int> & 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<int> & id_tasks, int timeout) {
while (true) {
std::unique_lock<std::mutex> 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<float> 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<int> & id_tasks,
const std::function<void(std::vector<server_task_result_ptr>&)> & result_handler,
const std::function<void(json)> & error_handler) {
const std::function<void(json)> & error_handler,
const std::function<bool()> & is_connection_closed) {
std::vector<server_task_result_ptr> 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<int> & id_tasks,
const std::function<bool(server_task_result_ptr&)> & result_handler,
const std::function<void(json)> & error_handler) {
const std::function<void(json)> & error_handler,
const std::function<bool()> & 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<bool()> 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<server_task> tasks;
try {
std::vector<llama_tokens> tokenized_prompts = tokenize_input_prompts(ctx_server.ctx, data.at("prompt"), true, true);
std::vector<llama_tokens> 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<llama_tokens> tokenized_prompts = tokenize_input_prompts(ctx_server.ctx, prompt, true, true);
std::vector<llama_tokens> 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<llama_tokens> tokenized_prompts = tokenize_input_prompts(ctx_server.ctx, prompt, true, true);
std::vector<llama_tokens> 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<server_task> tasks;
std::vector<llama_tokens> tokenized_docs = tokenize_input_prompts(ctx_server.ctx, documents, /* add_special */ false, true);
std::vector<llama_tokens> 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));

16
examples/server/tests/unit/test_chat_completion.py
View File

@ -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,

21
examples/server/tests/unit/test_completion.py
View File

@ -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

2
examples/server/tests/unit/test_infill.py
View File

@ -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():

19
examples/server/tests/utils.py
View File

@ -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()

121
examples/server/utils.hpp
View File

@ -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 <random>
#include <sstream>
@ -118,7 +120,7 @@ static json json_get_nested_values(const std::vector<std::string> & 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<std::string>();
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<llama_tokens> tokenize_input_prompts(llama_context * ctx, const json & json_prompt, bool add_special, bool parse_special) {
static std::vector<llama_tokens> tokenize_input_prompts(const llama_vocab * vocab, const json & json_prompt, bool add_special, bool parse_special) {
std::vector<llama_tokens> 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<llama_tokens>());
@ -179,7 +181,7 @@ static std::vector<llama_tokens> 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<llama_tokens>());
@ -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<json> & messages) {
inline std::string format_chat(const common_chat_template & tmpl, const std::vector<json> & messages) {
std::vector<common_chat_msg> 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<std::string> unsupported_params { "tools", "tool_choice" };
static const std::vector<std::string> 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<llama_logit_bias> & 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<llama_token_data> get_token_probabilities(llama_context * ctx, int idx) {
std::vector<llama_token_data> 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<llama_token_data> get_token_probabilities(llama_context * ctx
}
static bool are_lora_equal(
const std::vector<common_lora_adapter_info> & l1,
const std::vector<common_lora_adapter_info> & l2) {
const std::vector<common_adapter_lora_info> & l1,
const std::vector<common_adapter_lora_info> & 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<common_lora_adapter_info> parse_lora_request(
const std::vector<common_lora_adapter_info> & lora_base,
static std::vector<common_adapter_lora_info> parse_lora_request(
const std::vector<common_adapter_lora_info> & lora_base,
const json & data) {
std::vector<common_lora_adapter_info> lora(lora_base);
std::vector<common_adapter_lora_info> lora(lora_base);
int max_idx = lora.size();
// clear existing value

105
examples/server/webui/index.html
View File

@ -37,7 +37,7 @@
<div v-for="conv in conversations" :class="{
'btn btn-ghost justify-start font-normal': true,
'btn-active': conv.id === viewingConvId,
}" @click="setViewingConv(conv.id)">
}" @click="setViewingConv(conv.id)" dir="auto">
<span class="truncate">{{ conv.messages[0].content }}</span>
</div>
<div class="text-center text-xs opacity-40 mt-auto mx-4">
@ -62,53 +62,57 @@
<!-- action buttons (top right) -->
<div class="flex items-center">
<div v-if="messages.length > 0" class="dropdown dropdown-end">
<!-- "more" button -->
<!-- "..." button -->
<button tabindex="0" role="button" class="btn m-1" :disabled="isGenerating">
<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="currentColor" class="bi bi-three-dots-vertical" viewBox="0 0 16 16">
<path d="M9.5 13a1.5 1.5 0 1 1-3 0 1.5 1.5 0 0 1 3 0m0-5a1.5 1.5 0 1 1-3 0 1.5 1.5 0 0 1 3 0m0-5a1.5 1.5 0 1 1-3 0 1.5 1.5 0 0 1 3 0"/>
</svg>
</button>
<!-- "more" dropdown menu -->
<!-- "delete" dropdown menu -->
<ul tabindex="0" class="dropdown-content menu bg-base-100 rounded-box z-[1] w-52 p-2 shadow">
<li @click="downloadConv(viewingConvId)"><a>Download</a></li>
<li class="text-error" @click="deleteConv(viewingConvId)"><a>Delete</a></li>
</ul>
</div>
<button class="btn" @click="showConfigDialog = true" :disabled="isGenerating">
<!-- settings button -->
<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="currentColor" class="bi bi-gear" viewBox="0 0 16 16">
<path d="M8 4.754a3.246 3.246 0 1 0 0 6.492 3.246 3.246 0 0 0 0-6.492M5.754 8a2.246 2.246 0 1 1 4.492 0 2.246 2.246 0 0 1-4.492 0"/>
<path d="M9.796 1.343c-.527-1.79-3.065-1.79-3.592 0l-.094.319a.873.873 0 0 1-1.255.52l-.292-.16c-1.64-.892-3.433.902-2.54 2.541l.159.292a.873.873 0 0 1-.52 1.255l-.319.094c-1.79.527-1.79 3.065 0 3.592l.319.094a.873.873 0 0 1 .52 1.255l-.16.292c-.892 1.64.901 3.434 2.541 2.54l.292-.159a.873.873 0 0 1 1.255.52l.094.319c.527 1.79 3.065 1.79 3.592 0l.094-.319a.873.873 0 0 1 1.255-.52l.292.16c1.64.893 3.434-.902 2.54-2.541l-.159-.292a.873.873 0 0 1 .52-1.255l.319-.094c1.79-.527 1.79-3.065 0-3.592l-.319-.094a.873.873 0 0 1-.52-1.255l.16-.292c.893-1.64-.902-3.433-2.541-2.54l-.292.159a.873.873 0 0 1-1.255-.52zm-2.633.283c.246-.835 1.428-.835 1.674 0l.094.319a1.873 1.873 0 0 0 2.693 1.115l.291-.16c.764-.415 1.6.42 1.184 1.185l-.159.292a1.873 1.873 0 0 0 1.116 2.692l.318.094c.835.246.835 1.428 0 1.674l-.319.094a1.873 1.873 0 0 0-1.115 2.693l.16.291c.415.764-.42 1.6-1.185 1.184l-.291-.159a1.873 1.873 0 0 0-2.693 1.116l-.094.318c-.246.835-1.428.835-1.674 0l-.094-.319a1.873 1.873 0 0 0-2.692-1.115l-.292.16c-.764.415-1.6-.42-1.184-1.185l.159-.291A1.873 1.873 0 0 0 1.945 8.93l-.319-.094c-.835-.246-.835-1.428 0-1.674l.319-.094A1.873 1.873 0 0 0 3.06 4.377l-.16-.292c-.415-.764.42-1.6 1.185-1.184l.292.159a1.873 1.873 0 0 0 2.692-1.115z"/>
</svg>
</button>
<div class="tooltip tooltip-bottom" data-tip="Settings">
<button class="btn" @click="showConfigDialog = true" :disabled="isGenerating">
<!-- settings button -->
<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="currentColor" class="bi bi-gear" viewBox="0 0 16 16">
<path d="M8 4.754a3.246 3.246 0 1 0 0 6.492 3.246 3.246 0 0 0 0-6.492M5.754 8a2.246 2.246 0 1 1 4.492 0 2.246 2.246 0 0 1-4.492 0"/>
<path d="M9.796 1.343c-.527-1.79-3.065-1.79-3.592 0l-.094.319a.873.873 0 0 1-1.255.52l-.292-.16c-1.64-.892-3.433.902-2.54 2.541l.159.292a.873.873 0 0 1-.52 1.255l-.319.094c-1.79.527-1.79 3.065 0 3.592l.319.094a.873.873 0 0 1 .52 1.255l-.16.292c-.892 1.64.901 3.434 2.541 2.54l.292-.159a.873.873 0 0 1 1.255.52l.094.319c.527 1.79 3.065 1.79 3.592 0l.094-.319a.873.873 0 0 1 1.255-.52l.292.16c1.64.893 3.434-.902 2.54-2.541l-.159-.292a.873.873 0 0 1 .52-1.255l.319-.094c1.79-.527 1.79-3.065 0-3.592l-.319-.094a.873.873 0 0 1-.52-1.255l.16-.292c.893-1.64-.902-3.433-2.541-2.54l-.292.159a.873.873 0 0 1-1.255-.52zm-2.633.283c.246-.835 1.428-.835 1.674 0l.094.319a1.873 1.873 0 0 0 2.693 1.115l.291-.16c.764-.415 1.6.42 1.184 1.185l-.159.292a1.873 1.873 0 0 0 1.116 2.692l.318.094c.835.246.835 1.428 0 1.674l-.319.094a1.873 1.873 0 0 0-1.115 2.693l.16.291c.415.764-.42 1.6-1.185 1.184l-.291-.159a1.873 1.873 0 0 0-2.693 1.116l-.094.318c-.246.835-1.428.835-1.674 0l-.094-.319a1.873 1.873 0 0 0-2.692-1.115l-.292.16c-.764.415-1.6-.42-1.184-1.185l.159-.291A1.873 1.873 0 0 0 1.945 8.93l-.319-.094c-.835-.246-.835-1.428 0-1.674l.319-.094A1.873 1.873 0 0 0 3.06 4.377l-.16-.292c-.415-.764.42-1.6 1.185-1.184l.292.159a1.873 1.873 0 0 0 2.692-1.115z"/>
</svg>
</button>
</div>
<!-- theme controller is copied from https://daisyui.com/components/theme-controller/ -->
<div class="dropdown dropdown-end dropdown-bottom">
<div tabindex="0" role="button" class="btn m-1">
<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="currentColor" class="bi bi-palette2" viewBox="0 0 16 16">
<path d="M0 .5A.5.5 0 0 1 .5 0h5a.5.5 0 0 1 .5.5v5.277l4.147-4.131a.5.5 0 0 1 .707 0l3.535 3.536a.5.5 0 0 1 0 .708L10.261 10H15.5a.5.5 0 0 1 .5.5v5a.5.5 0 0 1-.5.5H3a3 3 0 0 1-2.121-.879A3 3 0 0 1 0 13.044m6-.21 7.328-7.3-2.829-2.828L6 7.188zM4.5 13a1.5 1.5 0 1 0-3 0 1.5 1.5 0 0 0 3 0M15 15v-4H9.258l-4.015 4zM0 .5v12.495zm0 12.495V13z"/>
</svg>
<div class="tooltip tooltip-bottom" data-tip="Themes">
<div class="dropdown dropdown-end dropdown-bottom">
<div tabindex="0" role="button" class="btn m-1">
<svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="currentColor" class="bi bi-palette2" viewBox="0 0 16 16">
<path d="M0 .5A.5.5 0 0 1 .5 0h5a.5.5 0 0 1 .5.5v5.277l4.147-4.131a.5.5 0 0 1 .707 0l3.535 3.536a.5.5 0 0 1 0 .708L10.261 10H15.5a.5.5 0 0 1 .5.5v5a.5.5 0 0 1-.5.5H3a3 3 0 0 1-2.121-.879A3 3 0 0 1 0 13.044m6-.21 7.328-7.3-2.829-2.828L6 7.188zM4.5 13a1.5 1.5 0 1 0-3 0 1.5 1.5 0 0 0 3 0M15 15v-4H9.258l-4.015 4zM0 .5v12.495zm0 12.495V13z"/>
</svg>
</div>
<ul tabindex="0" class="dropdown-content bg-base-300 rounded-box z-[1] w-52 p-2 shadow-2xl h-80 overflow-y-auto">
<li>
<button
class="btn btn-sm btn-block btn-ghost justify-start"
:class="{ 'btn-active': selectedTheme === 'auto' }"
@click="setSelectedTheme('auto')">
auto
</button>
</li>
<li v-for="theme in themes">
<input
type="radio"
name="theme-dropdown"
class="theme-controller btn btn-sm btn-block btn-ghost justify-start"
:aria-label="theme"
:value="theme"
:checked="selectedTheme === theme"
@click="setSelectedTheme(theme)" />
</li>
</ul>
</div>
<ul tabindex="0" class="dropdown-content bg-base-300 rounded-box z-[1] w-52 p-2 shadow-2xl h-80 overflow-y-auto">
<li>
<button
class="btn btn-sm btn-block btn-ghost justify-start"
:class="{ 'btn-active': selectedTheme === 'auto' }"
@click="setSelectedTheme('auto')">
auto
</button>
</li>
<li v-for="theme in themes">
<input
type="radio"
name="theme-dropdown"
class="theme-controller btn btn-sm btn-block btn-ghost justify-start"
:aria-label="theme"
:value="theme"
:checked="selectedTheme === theme"
@click="setSelectedTheme(theme)" />
</li>
</ul>
</div>
</div>
</div>
@ -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="() => {}"></message-bubble>
</div>
@ -152,6 +157,7 @@
@keydown.enter.shift.exact.prevent="inputMsg += '\n'"
:disabled="isGenerating"
id="msg-input"
dir="auto"
></textarea>
<button v-if="!isGenerating" class="btn btn-primary ml-2" @click="sendMessage" :disabled="inputMsg.length === 0">Send</button>
<button v-else class="btn btn-neutral ml-2" @click="stopGeneration">Stop</button>
@ -197,6 +203,20 @@
</template>
</div>
</details>
<!-- Section: Reasoning models -->
<details class="collapse collapse-arrow bg-base-200 mb-2 overflow-visible">
<summary class="collapse-title font-bold">Reasoning models</summary>
<div class="collapse-content">
<div class="flex flex-row items-center mb-2">
<input type="checkbox" class="checkbox" v-model="config.showThoughtInProgress" />
<span class="ml-4">Expand though process by default for generating message</span>
</div>
<div class="flex flex-row items-center mb-2">
<input type="checkbox" class="checkbox" v-model="config.excludeThoughtOnReq" />
<span class="ml-4">Exclude thought process when sending request to API (Recommended for DeepSeek-R1)</span>
</div>
</div>
</details>
<!-- Section: Advanced config -->
<details class="collapse collapse-arrow bg-base-200 mb-2 overflow-visible">
<summary class="collapse-title font-bold">Advanced config</summary>
@ -244,6 +264,7 @@
<!-- textarea for editing message -->
<template v-if="editingContent !== null">
<textarea
dir="auto"
class="textarea textarea-bordered bg-base-100 text-base-content w-[calc(90vw-8em)] lg:w-96"
v-model="editingContent"></textarea>
<br/>
@ -254,7 +275,19 @@
<!-- show loading dots for pending message -->
<span v-if="msg.content === null" class="loading loading-dots loading-md"></span>
<!-- render message as markdown -->
<vue-markdown v-else :source="msg.content"></vue-markdown>
<div v-else dir="auto">
<details v-if="msg.role === 'assistant' && splitMsgContent.cot" class="collapse bg-base-200 collapse-arrow mb-4" :open="splitMsgContent.isThinking && showThoughtInProgress">
<summary class="collapse-title">
<span v-if="splitMsgContent.isThinking">
<span v-if="isGenerating" class="loading loading-spinner loading-md mr-2" style="vertical-align: middle;"></span>
<b>Thinking</b>
</span>
<b v-else>Thought Process</b>
</summary>
<vue-markdown :source="splitMsgContent.cot" dir="auto" class="collapse-content"></vue-markdown>
</details>
<vue-markdown :source="splitMsgContent.content"></vue-markdown>
</div>
<!-- render timings if enabled -->
<div class="dropdown dropdown-hover dropdown-top mt-2" v-if="timings && config.showTokensPerSecond">
<div tabindex="0" role="button" class="cursor-pointer font-semibold text-sm opacity-60">Speed: {{ timings.predicted_per_second.toFixed(1) }} t/s</div>

169
examples/server/webui/src/main.js
View File

@ -17,6 +17,11 @@ import { asyncIterator } from '@sec-ant/readable-stream/ponyfill/asyncIterator';
const isDev = import.meta.env.MODE === 'development';
// types
/** @typedef {{ id: number, role: 'user' | 'assistant', content: string, timings: any }} Message */
/** @typedef {{ role: 'user' | 'assistant', content: string }} APIMessage */
/** @typedef {{ id: string, lastModified: number, messages: Array<Message> }} Conversation */
// utility functions
const isString = (x) => !!x.toLowerCase;
const isBoolean = (x) => x === true || x === false;
@ -50,6 +55,8 @@ const CONFIG_DEFAULT = {
apiKey: '',
systemMessage: 'You are a helpful assistant.',
showTokensPerSecond: false,
showThoughtInProgress: false,
excludeThoughtOnReq: true,
// make sure these default values are in sync with `common.h`
samplers: 'edkypmxt',
temperature: 0.8,
@ -111,12 +118,12 @@ const VueMarkdown = defineComponent(
highlight: function (str, lang) { // Add highlight.js
if (lang && hljs.getLanguage(lang)) {
try {
return '<pre><code class="hljs">' +
return '<pre dir="auto"><code class="hljs">' +
hljs.highlight(str, { language: lang, ignoreIllegals: true }).value +
'</code></pre>';
} catch (__) {}
}
return '<pre><code class="hljs">' + md.value.utils.escapeHtml(str) + '</code></pre>';
return '<pre dir="auto"><code class="hljs">' + md.value.utils.escapeHtml(str) + '</code></pre>';
}
}));
// support latex with double dollar sign and square brackets
@ -172,6 +179,7 @@ const MessageBubble = defineComponent({
config: Object,
msg: Object,
isGenerating: Boolean,
showThoughtInProgress: Boolean,
editUserMsgAndRegenerate: Function,
regenerateMsg: Function,
},
@ -188,7 +196,31 @@ const MessageBubble = defineComponent({
prompt_per_second: this.msg.timings.prompt_n / (this.msg.timings.prompt_ms / 1000),
predicted_per_second: this.msg.timings.predicted_n / (this.msg.timings.predicted_ms / 1000),
};
}
},
splitMsgContent() {
const content = this.msg.content;
if (this.msg.role !== 'assistant') {
return { content };
}
let actualContent = '';
let cot = '';
let isThinking = false;
let thinkSplit = content.split('<think>', 2);
actualContent += thinkSplit[0];
while (thinkSplit[1] !== undefined) {
// <think> tag found
thinkSplit = thinkSplit[1].split('</think>', 2);
cot += thinkSplit[0];
isThinking = true;
if (thinkSplit[1] !== undefined) {
// </think> closing tag found
isThinking = false;
thinkSplit = thinkSplit[1].split('<think>', 2);
actualContent += thinkSplit[0];
}
}
return { content: actualContent, cot, isThinking };
},
},
methods: {
copyMsg() {
@ -208,7 +240,10 @@ const MessageBubble = defineComponent({
// format: { [convId]: { id: string, lastModified: number, messages: [...] } }
// convId is a string prefixed with 'conv-'
const StorageUtils = {
// manage conversations
/**
* manage conversations
* @returns {Array<Conversation>}
*/
getAllConversations() {
const res = [];
for (const key in localStorage) {
@ -219,11 +254,19 @@ const StorageUtils = {
res.sort((a, b) => b.lastModified - a.lastModified);
return res;
},
// can return null if convId does not exist
/**
* can return null if convId does not exist
* @param {string} convId
* @returns {Conversation | null}
*/
getOneConversation(convId) {
return JSON.parse(localStorage.getItem(convId) || 'null');
},
// if convId does not exist, create one
/**
* if convId does not exist, create one
* @param {string} convId
* @param {Message} msg
*/
appendMsg(convId, msg) {
if (msg.content === null) return;
const conv = StorageUtils.getOneConversation(convId) || {
@ -235,12 +278,24 @@ const StorageUtils = {
conv.lastModified = Date.now();
localStorage.setItem(convId, JSON.stringify(conv));
},
/**
* Get new conversation id
* @returns {string}
*/
getNewConvId() {
return `conv-${Date.now()}`;
},
/**
* remove conversation by id
* @param {string} convId
*/
remove(convId) {
localStorage.removeItem(convId);
},
/**
* remove all conversations
* @param {string} convId
*/
filterAndKeepMsgs(convId, predicate) {
const conv = StorageUtils.getOneConversation(convId);
if (!conv) return;
@ -248,6 +303,11 @@ const StorageUtils = {
conv.lastModified = Date.now();
localStorage.setItem(convId, JSON.stringify(conv));
},
/**
* remove last message from conversation
* @param {string} convId
* @returns {Message | undefined}
*/
popMsg(convId) {
const conv = StorageUtils.getOneConversation(convId);
if (!conv) return;
@ -322,10 +382,12 @@ const mainApp = createApp({
data() {
return {
conversations: StorageUtils.getAllConversations(),
messages: [], // { id: number, role: 'user' | 'assistant', content: string }
/** @type {Array<Message>} */
messages: [],
viewingConvId: StorageUtils.getNewConvId(),
inputMsg: '',
isGenerating: false,
/** @type {Array<Message> | null} */
pendingMsg: null, // the on-going message from assistant
stopGeneration: () => {},
selectedTheme: StorageUtils.getTheme(),
@ -333,6 +395,7 @@ const mainApp = createApp({
showConfigDialog: false,
// const
themes: THEMES,
/** @type {CONFIG_DEFAULT} */
configDefault: {...CONFIG_DEFAULT},
configInfo: {...CONFIG_INFO},
isDev,
@ -425,42 +488,50 @@ const mainApp = createApp({
this.isGenerating = true;
try {
/** @type {CONFIG_DEFAULT} */
const config = this.config;
const abortController = new AbortController();
this.stopGeneration = () => abortController.abort();
/** @type {Array<APIMessage>} */
let messages = [
{ role: 'system', content: config.systemMessage },
...normalizeMsgsForAPI(this.messages),
];
if (config.excludeThoughtOnReq) {
messages = filterThoughtFromMsgs(messages);
}
if (isDev) console.log({messages});
const params = {
messages: [
{ role: 'system', content: this.config.systemMessage },
...this.messages,
],
messages,
stream: true,
cache_prompt: true,
samplers: this.config.samplers,
temperature: this.config.temperature,
dynatemp_range: this.config.dynatemp_range,
dynatemp_exponent: this.config.dynatemp_exponent,
top_k: this.config.top_k,
top_p: this.config.top_p,
min_p: this.config.min_p,
typical_p: this.config.typical_p,
xtc_probability: this.config.xtc_probability,
xtc_threshold: this.config.xtc_threshold,
repeat_last_n: this.config.repeat_last_n,
repeat_penalty: this.config.repeat_penalty,
presence_penalty: this.config.presence_penalty,
frequency_penalty: this.config.frequency_penalty,
dry_multiplier: this.config.dry_multiplier,
dry_base: this.config.dry_base,
dry_allowed_length: this.config.dry_allowed_length,
dry_penalty_last_n: this.config.dry_penalty_last_n,
max_tokens: this.config.max_tokens,
timings_per_token: !!this.config.showTokensPerSecond,
...(this.config.custom.length ? JSON.parse(this.config.custom) : {}),
samplers: config.samplers,
temperature: config.temperature,
dynatemp_range: config.dynatemp_range,
dynatemp_exponent: config.dynatemp_exponent,
top_k: config.top_k,
top_p: config.top_p,
min_p: config.min_p,
typical_p: config.typical_p,
xtc_probability: config.xtc_probability,
xtc_threshold: config.xtc_threshold,
repeat_last_n: config.repeat_last_n,
repeat_penalty: config.repeat_penalty,
presence_penalty: config.presence_penalty,
frequency_penalty: config.frequency_penalty,
dry_multiplier: config.dry_multiplier,
dry_base: config.dry_base,
dry_allowed_length: config.dry_allowed_length,
dry_penalty_last_n: config.dry_penalty_last_n,
max_tokens: config.max_tokens,
timings_per_token: !!config.showTokensPerSecond,
...(config.custom.length ? JSON.parse(config.custom) : {}),
};
const chunks = sendSSEPostRequest(`${BASE_URL}/v1/chat/completions`, {
method: 'POST',
headers: {
'Content-Type': 'application/json',
...(this.config.apiKey ? {'Authorization': `Bearer ${this.config.apiKey}`} : {})
...(config.apiKey ? {'Authorization': `Bearer ${config.apiKey}`} : {})
},
body: JSON.stringify(params),
signal: abortController.signal,
@ -477,7 +548,7 @@ const mainApp = createApp({
};
}
const timings = chunk.timings;
if (timings && this.config.showTokensPerSecond) {
if (timings && config.showTokensPerSecond) {
// only extract what's really needed, to save some space
this.pendingMsg.timings = {
prompt_n: timings.prompt_n,
@ -598,3 +669,33 @@ try {
<button class="btn" onClick="localStorage.clear(); window.location.reload();">Clear localStorage</button>
</div>`;
}
/**
* filter out redundant fields upon sending to API
* @param {Array<APIMessage>} messages
* @returns {Array<APIMessage>}
*/
function normalizeMsgsForAPI(messages) {
return messages.map((msg) => {
return {
role: msg.role,
content: msg.content,
};
});
}
/**
* recommended for DeepsSeek-R1, filter out content between <think> and </think> tags
* @param {Array<APIMessage>} messages
* @returns {Array<APIMessage>}
*/
function filterThoughtFromMsgs(messages) {
return messages.map((msg) => {
return {
role: msg.role,
content: msg.role === 'assistant'
? msg.content.split('</think>').at(-1).trim()
: msg.content,
};
});
}

26
examples/simple-chat/simple-chat.cpp
View File

@ -69,18 +69,20 @@ int main(int argc, char ** argv) {
llama_model_params model_params = llama_model_default_params();
model_params.n_gpu_layers = ngl;
llama_model * model = llama_load_model_from_file(model_path.c_str(), model_params);
llama_model * model = llama_model_load_from_file(model_path.c_str(), model_params);
if (!model) {
fprintf(stderr , "%s: error: 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 = llama_context_default_params();
ctx_params.n_ctx = n_ctx;
ctx_params.n_batch = n_ctx;
llama_context * ctx = llama_new_context_with_model(model, ctx_params);
llama_context * ctx = llama_init_from_model(model, ctx_params);
if (!ctx) {
fprintf(stderr , "%s: error: failed to create the llama_context\n" , __func__);
return 1;
@ -96,10 +98,12 @@ int main(int argc, char ** argv) {
auto generate = [&](const std::string & prompt) {
std::string response;
const bool is_first = llama_get_kv_cache_used_cells(ctx) == 0;
// tokenize the prompt
const int n_prompt_tokens = -llama_tokenize(model, prompt.c_str(), prompt.size(), NULL, 0, true, true);
const int n_prompt_tokens = -llama_tokenize(vocab, prompt.c_str(), prompt.size(), NULL, 0, is_first, true);
std::vector<llama_token> prompt_tokens(n_prompt_tokens);
if (llama_tokenize(model, prompt.c_str(), prompt.size(), prompt_tokens.data(), prompt_tokens.size(), llama_get_kv_cache_used_cells(ctx) == 0, true) < 0) {
if (llama_tokenize(vocab, prompt.c_str(), prompt.size(), prompt_tokens.data(), prompt_tokens.size(), is_first, true) < 0) {
GGML_ABORT("failed to tokenize the prompt\n");
}
@ -124,13 +128,13 @@ int main(int argc, char ** argv) {
new_token_id = llama_sampler_sample(smpl, ctx, -1);
// is it an end of generation?
if (llama_token_is_eog(model, new_token_id)) {
if (llama_vocab_is_eog(vocab, new_token_id)) {
break;
}
// convert the token to a string, print it and add it to the response
char buf[256];
int n = llama_token_to_piece(model, new_token_id, buf, sizeof(buf), 0, true);
int n = llama_token_to_piece(vocab, new_token_id, buf, sizeof(buf), 0, true);
if (n < 0) {
GGML_ABORT("failed to convert token to piece\n");
}
@ -159,12 +163,14 @@ int main(int argc, char ** argv) {
break;
}
const char * tmpl = llama_model_chat_template(model, /* name */ nullptr);
// add the user input to the message list and format it
messages.push_back({"user", strdup(user.c_str())});
int new_len = llama_chat_apply_template(model, nullptr, messages.data(), messages.size(), true, formatted.data(), formatted.size());
int new_len = llama_chat_apply_template(tmpl, messages.data(), messages.size(), true, formatted.data(), formatted.size());
if (new_len > (int)formatted.size()) {
formatted.resize(new_len);
new_len = llama_chat_apply_template(model, nullptr, messages.data(), messages.size(), true, formatted.data(), formatted.size());
new_len = llama_chat_apply_template(tmpl, messages.data(), messages.size(), true, formatted.data(), formatted.size());
}
if (new_len < 0) {
fprintf(stderr, "failed to apply the chat template\n");
@ -181,7 +187,7 @@ int main(int argc, char ** argv) {
// add the response to the messages
messages.push_back({"assistant", strdup(response.c_str())});
prev_len = llama_chat_apply_template(model, nullptr, messages.data(), messages.size(), false, nullptr, 0);
prev_len = llama_chat_apply_template(tmpl, messages.data(), messages.size(), false, nullptr, 0);
if (prev_len < 0) {
fprintf(stderr, "failed to apply the chat template\n");
return 1;
@ -194,7 +200,7 @@ int main(int argc, char ** argv) {
}
llama_sampler_free(smpl);
llama_free(ctx);
llama_free_model(model);
llama_model_free(model);
return 0;
}

17
examples/simple/simple.cpp
View File

@ -83,7 +83,8 @@ int main(int argc, char ** argv) {
llama_model_params model_params = llama_model_default_params();
model_params.n_gpu_layers = ngl;
llama_model * model = llama_load_model_from_file(model_path.c_str(), model_params);
llama_model * model = llama_model_load_from_file(model_path.c_str(), model_params);
const llama_vocab * vocab = llama_model_get_vocab(model);
if (model == NULL) {
fprintf(stderr , "%s: error: unable to load model\n" , __func__);
@ -93,11 +94,11 @@ int main(int argc, char ** argv) {
// tokenize the prompt
// find the number of tokens in the prompt
const int n_prompt = -llama_tokenize(model, prompt.c_str(), prompt.size(), NULL, 0, true, true);
const int n_prompt = -llama_tokenize(vocab, prompt.c_str(), prompt.size(), NULL, 0, true, true);
// allocate space for the tokens and tokenize the prompt
std::vector<llama_token> prompt_tokens(n_prompt);
if (llama_tokenize(model, prompt.c_str(), prompt.size(), prompt_tokens.data(), prompt_tokens.size(), true, true) < 0) {
if (llama_tokenize(vocab, prompt.c_str(), prompt.size(), prompt_tokens.data(), prompt_tokens.size(), true, true) < 0) {
fprintf(stderr, "%s: error: failed to tokenize the prompt\n", __func__);
return 1;
}
@ -112,7 +113,7 @@ int main(int argc, char ** argv) {
// enable performance counters
ctx_params.no_perf = false;
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__);
@ -131,7 +132,7 @@ int main(int argc, char ** argv) {
for (auto id : prompt_tokens) {
char buf[128];
int n = llama_token_to_piece(model, id, buf, sizeof(buf), 0, true);
int n = llama_token_to_piece(vocab, id, buf, sizeof(buf), 0, true);
if (n < 0) {
fprintf(stderr, "%s: error: failed to convert token to piece\n", __func__);
return 1;
@ -164,12 +165,12 @@ int main(int argc, char ** argv) {
new_token_id = llama_sampler_sample(smpl, ctx, -1);
// is it an end of generation?
if (llama_token_is_eog(model, new_token_id)) {
if (llama_vocab_is_eog(vocab, new_token_id)) {
break;
}
char buf[128];
int n = llama_token_to_piece(model, new_token_id, buf, sizeof(buf), 0, true);
int n = llama_token_to_piece(vocab, new_token_id, buf, sizeof(buf), 0, true);
if (n < 0) {
fprintf(stderr, "%s: error: failed to convert token to piece\n", __func__);
return 1;
@ -199,7 +200,7 @@ int main(int argc, char ** argv) {
llama_sampler_free(smpl);
llama_free(ctx);
llama_free_model(model);
llama_model_free(model);
return 0;
}

4
examples/speculative-simple/speculative-simple.cpp
View File

@ -45,6 +45,8 @@ int main(int argc, char ** argv) {
model_tgt = llama_init_tgt.model.get();
ctx_tgt = llama_init_tgt.context.get();
const llama_vocab * vocab = llama_model_get_vocab(model_tgt);
// load the draft model
params.devices = params.speculative.devices;
params.model = params.speculative.model;
@ -196,7 +198,7 @@ int main(int argc, char ** argv) {
id_last = ids[i];
if (llama_token_is_eog(model_tgt, id_last)) {
if (llama_vocab_is_eog(vocab, id_last)) {
has_eos = true;
break;
}

29
examples/speculative/speculative.cpp
View File

@ -90,10 +90,13 @@ int main(int argc, char ** argv) {
model_dft = llama_init_dft.model.get();
ctx_dft = llama_init_dft.context.get();
const bool vocab_type_tgt = llama_vocab_type(model_tgt);
const llama_vocab * vocab_tgt = llama_model_get_vocab(model_tgt);
const llama_vocab * vocab_dft = llama_model_get_vocab(model_dft);
const bool vocab_type_tgt = llama_vocab_type(vocab_tgt);
LOG_DBG("vocab_type tgt: %d\n", 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("vocab_type dft: %d\n", vocab_type_dft);
if (vocab_type_tgt != vocab_type_dft) {
@ -103,18 +106,18 @@ int main(int argc, char ** argv) {
}
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)
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 model special tokens must match target model to use speculation\n", __func__);
return 1;
}
{
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 = n_vocab_tgt > n_vocab_dft
? n_vocab_tgt - n_vocab_dft
: n_vocab_dft - n_vocab_tgt;
@ -122,13 +125,13 @@ int main(int argc, char ** argv) {
if (vocab_diff > SPEC_VOCAB_MAX_SIZE_DIFFERENCE) {
LOG_ERR("%s: draft model vocab must closely match target model to use speculation but ", __func__);
LOG_ERR("target vocab size %d does not match draft vocab size %d - difference %d, max allowed %d\n",
n_vocab_tgt, llama_n_vocab(model_dft), vocab_diff, SPEC_VOCAB_MAX_SIZE_DIFFERENCE);
n_vocab_tgt, llama_vocab_n_tokens(vocab_dft), vocab_diff, SPEC_VOCAB_MAX_SIZE_DIFFERENCE);
return 1;
}
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 ", __func__);
LOG_ERR("token %d content differs - target '%s', draft '%s'\n", i,
@ -170,7 +173,7 @@ int main(int argc, char ** argv) {
const auto t_enc_end = ggml_time_us();
// the 2 models should have the same vocab
//GGML_ASSERT(n_vocab == llama_n_vocab(model_dft));
//GGML_ASSERT(n_vocab == llama_vocab_n_tokens(model_dft));
// how many tokens to draft each time
int n_draft = params.speculative.n_max;
@ -386,7 +389,7 @@ int main(int argc, char ** argv) {
}
}
if (llama_token_is_eog(model_tgt, token_id)) {
if (llama_vocab_is_eog(vocab_tgt, token_id)) {
has_eos = true;
}
++n_predict;

22
examples/tokenize/tokenize.cpp
View File

@ -31,6 +31,7 @@ static void print_usage_information(const char * argv0) {
printf(" -p PROMPT, --prompt PROMPT read prompt from the argument.\n");
printf(" --stdin read prompt from standard input.\n");
printf(" --no-bos do not ever add a BOS token to the prompt, even if normally the model uses a BOS token.\n");
printf(" --no-escape do not escape input (such as \\n, \\t, etc.).\n");
printf(" --no-parse-special do not parse control tokens.\n");
printf(" --log-disable disable logs. Makes stderr quiet when loading the model.\n");
printf(" --show-count print the total number of tokens.\n");
@ -198,6 +199,7 @@ int main(int raw_argc, char ** raw_argv) {
// variables where to put any arguments we see.
bool printing_ids = false;
bool no_bos = false;
bool no_escape = false;
bool no_parse_special = false;
bool disable_logging = false;
bool show_token_count = false;
@ -233,6 +235,9 @@ int main(int raw_argc, char ** raw_argv) {
else if (arg == "--no-bos") {
no_bos = true;
}
else if (arg == "--no-escape") {
no_escape = true;
}
else if (arg == "--no-parse-special") {
no_parse_special = true;
}
@ -333,14 +338,16 @@ int main(int raw_argc, char ** raw_argv) {
llama_model_params model_params = llama_model_default_params();
model_params.vocab_only = true;
llama_model * model = llama_load_model_from_file(model_path, model_params);
llama_model * model = llama_model_load_from_file(model_path, model_params);
if (!model) {
fprintf(stderr, "Error: could not load model from file '%s'.\n", model_path);
return 1;
}
const llama_vocab * vocab = llama_model_get_vocab(model);
llama_context_params ctx_params = llama_context_default_params();
llama_context * ctx = llama_new_context_with_model(model, ctx_params);
llama_context * ctx = llama_init_from_model(model, ctx_params);
if (!ctx) {
fprintf(stderr, "Error: could not create context.\n");
return 1;
@ -360,12 +367,17 @@ int main(int raw_argc, char ** raw_argv) {
prompt = stdin_buffer.str();
}
const bool model_wants_add_bos = llama_add_bos_token(model);
const bool model_wants_add_bos = llama_vocab_get_add_bos(vocab);
const bool add_bos = model_wants_add_bos && !no_bos;
const bool parse_special = !no_parse_special;
const bool escape = !no_escape;
if (escape) {
string_process_escapes(prompt);
}
std::vector<llama_token> tokens;
tokens = common_tokenize(model, prompt, add_bos, parse_special);
tokens = common_tokenize(vocab, prompt, add_bos, parse_special);
if (printing_ids) {
printf("[");
@ -398,7 +410,7 @@ int main(int raw_argc, char ** raw_argv) {
}
// silence valgrind
llama_free(ctx);
llama_free_model(model);
llama_model_free(model);
return 0;
}

117
examples/tts/README.md Normal file
View File

@ -0,0 +1,117 @@
# llama.cpp/example/tts
This example demonstrates the Text To Speech feature. It uses a
[model](https://www.outeai.com/blog/outetts-0.2-500m) from
[outeai](https://www.outeai.com/).
## Quickstart
If you have built llama.cpp with `-DLLAMA_CURL=ON` you can simply run the
following command and the required models will be downloaded automatically:
```console
$ build/bin/llama-tts --tts-oute-default -p "Hello world" && aplay output.wav
```
For details about the models and how to convert them to the required format
see the following sections.
### Model conversion
Checkout or download the model that contains the LLM model:
```console
$ pushd models
$ git clone --branch main --single-branch --depth 1 https://huggingface.co/OuteAI/OuteTTS-0.2-500M
$ cd OuteTTS-0.2-500M && git lfs install && git lfs pull
$ popd
```
Convert the model to .gguf format:
```console
(venv) python convert_hf_to_gguf.py models/OuteTTS-0.2-500M \
--outfile models/outetts-0.2-0.5B-f16.gguf --outtype f16
```
The generated model will be `models/outetts-0.2-0.5B-f16.gguf`.
We can optionally quantize this to Q8_0 using the following command:
```console
$ build/bin/llama-quantize models/outetts-0.2-0.5B-f16.gguf \
models/outetts-0.2-0.5B-q8_0.gguf q8_0
```
The quantized model will be `models/outetts-0.2-0.5B-q8_0.gguf`.
Next we do something simlar for the audio decoder. First download or checkout
the model for the voice decoder:
```console
$ pushd models
$ git clone --branch main --single-branch --depth 1 https://huggingface.co/novateur/WavTokenizer-large-speech-75token
$ cd WavTokenizer-large-speech-75token && git lfs install && git lfs pull
$ popd
```
This model file is PyTorch checkpoint (.ckpt) and we first need to convert it to
huggingface format:
```console
(venv) python examples/tts/convert_pt_to_hf.py \
models/WavTokenizer-large-speech-75token/wavtokenizer_large_speech_320_24k.ckpt
...
Model has been successfully converted and saved to models/WavTokenizer-large-speech-75token/model.safetensors
Metadata has been saved to models/WavTokenizer-large-speech-75token/index.json
Config has been saved to models/WavTokenizer-large-speech-75tokenconfig.json
```
Then we can convert the huggingface format to gguf:
```console
(venv) python convert_hf_to_gguf.py models/WavTokenizer-large-speech-75token \
--outfile models/wavtokenizer-large-75-f16.gguf --outtype f16
...
INFO:hf-to-gguf:Model successfully exported to models/wavtokenizer-large-75-f16.gguf
```
### Running the example
With both of the models generated, the LLM model and the voice decoder model,
we can run the example:
```console
$ build/bin/llama-tts -m ./models/outetts-0.2-0.5B-q8_0.gguf \
-mv ./models/wavtokenizer-large-75-f16.gguf \
-p "Hello world"
...
main: audio written to file 'output.wav'
```
The output.wav file will contain the audio of the prompt. This can be heard
by playing the file with a media player. On Linux the following command will
play the audio:
```console
$ aplay output.wav
```
### Running the example with llama-server
Running this example with `llama-server` is also possible and requires two
server instances to be started. One will serve the LLM model and the other
will serve the voice decoder model.
The LLM model server can be started with the following command:
```console
$ ./build/bin/llama-server -m ./models/outetts-0.2-0.5B-q8_0.gguf --port 8020
```
And the voice decoder model server can be started using:
```console
./build/bin/llama-server -m ./models/wavtokenizer-large-75-f16.gguf --port 8021 --embeddings --pooling none
```
Then we can run [tts-outetts.py](tts-outetts.py) to generate the audio.
First create a virtual environment for python and install the required
dependencies (this in only required to be done once):
```console
$ python3 -m venv venv
$ source venv/bin/activate
(venv) pip install requests numpy
```
And then run the python script using:
```conole
(venv) python ./examples/tts/tts-outetts.py http://localhost:8020 http://localhost:8021 "Hello world"
spectrogram generated: n_codes: 90, n_embd: 1282
converting to audio ...
audio generated: 28800 samples
audio written to file "output.wav"
```
And to play the audio we can again use aplay or any other media player:
```console
$ aplay output.wav
```

128
examples/tts/tts-outetts.py
View File

@ -3,6 +3,121 @@ import sys
#import struct
import requests
import re
import struct
import numpy as np
from concurrent.futures import ThreadPoolExecutor
def fill_hann_window(size, periodic=True):
if periodic:
return np.hanning(size + 1)[:-1]
return np.hanning(size)
def irfft(n_fft, complex_input):
return np.fft.irfft(complex_input, n=n_fft)
def fold(buffer, n_out, n_win, n_hop, n_pad):
result = np.zeros(n_out)
n_frames = len(buffer) // n_win
for i in range(n_frames):
start = i * n_hop
end = start + n_win
result[start:end] += buffer[i * n_win:(i + 1) * n_win]
return result[n_pad:-n_pad] if n_pad > 0 else result
def process_frame(args):
l, n_fft, ST, hann = args
frame = irfft(n_fft, ST[l])
frame = frame * hann
hann2 = hann * hann
return frame, hann2
def embd_to_audio(embd, n_codes, n_embd, n_thread=4):
embd = np.asarray(embd, dtype=np.float32).reshape(n_codes, n_embd)
n_fft = 1280
n_hop = 320
n_win = 1280
n_pad = (n_win - n_hop) // 2
n_out = (n_codes - 1) * n_hop + n_win
hann = fill_hann_window(n_fft, True)
E = np.zeros((n_embd, n_codes), dtype=np.float32)
for l in range(n_codes):
for k in range(n_embd):
E[k, l] = embd[l, k]
half_embd = n_embd // 2
S = np.zeros((n_codes, half_embd + 1), dtype=np.complex64)
for k in range(half_embd):
for l in range(n_codes):
mag = E[k, l]
phi = E[k + half_embd, l]
mag = np.clip(np.exp(mag), 0, 1e2)
S[l, k] = mag * np.exp(1j * phi)
res = np.zeros(n_codes * n_fft)
hann2_buffer = np.zeros(n_codes * n_fft)
with ThreadPoolExecutor(max_workers=n_thread) as executor:
args = [(l, n_fft, S, hann) for l in range(n_codes)]
results = list(executor.map(process_frame, args))
for l, (frame, hann2) in enumerate(results):
res[l*n_fft:(l+1)*n_fft] = frame
hann2_buffer[l*n_fft:(l+1)*n_fft] = hann2
audio = fold(res, n_out, n_win, n_hop, n_pad)
env = fold(hann2_buffer, n_out, n_win, n_hop, n_pad)
mask = env > 1e-10
audio[mask] /= env[mask]
return audio
def save_wav(filename, audio_data, sample_rate):
num_channels = 1
bits_per_sample = 16
bytes_per_sample = bits_per_sample // 8
data_size = len(audio_data) * bytes_per_sample
byte_rate = sample_rate * num_channels * bytes_per_sample
block_align = num_channels * bytes_per_sample
chunk_size = 36 + data_size # 36 = size of header minus first 8 bytes
header = struct.pack(
'<4sI4s4sIHHIIHH4sI',
b'RIFF',
chunk_size,
b'WAVE',
b'fmt ',
16, # fmt chunk size
1, # audio format (PCM)
num_channels,
sample_rate,
byte_rate,
block_align,
bits_per_sample,
b'data',
data_size
)
audio_data = np.clip(audio_data * 32767, -32768, 32767)
pcm_data = audio_data.astype(np.int16)
with open(filename, 'wb') as f:
f.write(header)
f.write(pcm_data.tobytes())
def process_text(text: str):
text = re.sub(r'\d+(\.\d+)?', lambda x: x.group(), text.lower()) # TODO this needs to be fixed
@ -170,6 +285,15 @@ n_embd = len(embd[0])
print('spectrogram generated: n_codes: %d, n_embd: %d' % (n_codes, n_embd))
# post-process the spectrogram to convert to audio
# TODO: see the tts.cpp:embd_to_audio() and implement it in Python
print('converting to audio ...')
print('TODO: see the tts.cpp:embd_to_audio() and implement it in Python')
audio = embd_to_audio(embd, n_codes, n_embd)
print('audio generated: %d samples' % len(audio))
filename = "output.wav"
sample_rate = 24000 # sampling rate
# zero out first 0.25 seconds
audio[:24000 // 4] = 0.0
save_wav(filename, audio, sample_rate)
print('audio written to file "%s"' % filename)

71
examples/tts/tts.cpp
View File

@ -414,15 +414,42 @@ static void prompt_add(llama_tokens & prompt, const llama_tokens & tokens) {
prompt.insert(prompt.end(), tokens.begin(), tokens.end());
}
static void prompt_add(llama_tokens & prompt, const llama_model * model, const std::string & txt, bool add_special, bool parse_special) {
auto tmp = common_tokenize(model, txt, add_special, parse_special);
static void prompt_add(llama_tokens & prompt, const llama_vocab * vocab, const std::string & txt, bool add_special, bool parse_special) {
auto tmp = common_tokenize(vocab, txt, add_special, parse_special);
prompt_add(prompt, tmp);
}
static void prompt_init(llama_tokens & prompt, const llama_model * model) {
static void prompt_init(llama_tokens & prompt, const llama_vocab * vocab) {
prompt.clear();
prompt_add(prompt, model, "<|im_start|>\n", true, true);
prompt_add(prompt, vocab, "<|im_start|>\n", true, true);
}
static std::vector<llama_token> prepare_guide_tokens(const llama_vocab * vocab, const std::string & str) {
const std::string& delimiter = "<|text_sep|>";
std::vector<llama_token> result;
size_t start = 0;
size_t end = str.find(delimiter);
//first token is always a newline, as it was not previously added
result.push_back(common_tokenize(vocab, "\n", false, true)[0]);
while (end != std::string::npos) {
std::string current_word = str.substr(start, end - start);
auto tmp = common_tokenize(vocab, current_word, false, true);
result.push_back(tmp[0]);
start = end + delimiter.length();
end = str.find(delimiter, start);
}
// Add the last part
std::string current_word = str.substr(start);
auto tmp = common_tokenize(vocab, current_word, false, true);
if (tmp.size() > 0) {
result.push_back(tmp[0]);
}
return result;
}
int main(int argc, char ** argv) {
@ -462,6 +489,8 @@ int main(int argc, char ** argv) {
model_ttc = llama_init_ttc.model.get();
ctx_ttc = llama_init_ttc.context.get();
const llama_vocab * vocab = llama_model_get_vocab(model_ttc);
// TODO: refactor in a common struct
params.model = params.vocoder.model;
params.model_url = params.vocoder.model_url;
@ -492,6 +521,7 @@ int main(int argc, char ** argv) {
const auto t_main_start = ggml_time_us();
std::vector<llama_token> codes;
std::vector<llama_token> guide_tokens;
// process prompt and generate voice codes
{
@ -499,20 +529,23 @@ int main(int argc, char ** argv) {
std::vector<llama_token> prompt_inp;
prompt_init(prompt_inp, model_ttc);
prompt_init(prompt_inp, vocab);
prompt_add(prompt_inp, model_ttc, "<|text_start|>the<|text_sep|>overall<|text_sep|>package<|text_sep|>from<|text_sep|>just<|text_sep|>two<|text_sep|>people<|text_sep|>is<|text_sep|>pretty<|text_sep|>remarkable<|text_sep|>sure<|text_sep|>i<|text_sep|>have<|text_sep|>some<|text_sep|>critiques<|text_sep|>about<|text_sep|>some<|text_sep|>of<|text_sep|>the<|text_sep|>gameplay<|text_sep|>aspects<|text_sep|>but<|text_sep|>its<|text_sep|>still<|text_sep|>really<|text_sep|>enjoyable<|text_sep|>and<|text_sep|>it<|text_sep|>looks<|text_sep|>lovely<|text_sep|>", false, true);
prompt_add(prompt_inp, vocab, "<|text_start|>the<|text_sep|>overall<|text_sep|>package<|text_sep|>from<|text_sep|>just<|text_sep|>two<|text_sep|>people<|text_sep|>is<|text_sep|>pretty<|text_sep|>remarkable<|text_sep|>sure<|text_sep|>i<|text_sep|>have<|text_sep|>some<|text_sep|>critiques<|text_sep|>about<|text_sep|>some<|text_sep|>of<|text_sep|>the<|text_sep|>gameplay<|text_sep|>aspects<|text_sep|>but<|text_sep|>its<|text_sep|>still<|text_sep|>really<|text_sep|>enjoyable<|text_sep|>and<|text_sep|>it<|text_sep|>looks<|text_sep|>lovely<|text_sep|>", false, true);
// convert the input text into the necessary format expected by OuteTTS
{
std::string prompt_clean = process_text(params.prompt);
if (params.vocoder.use_guide_tokens) {
guide_tokens = prepare_guide_tokens(vocab, prompt_clean);
}
LOG_INF("%s: prompt: '%s'\n", __func__, prompt_clean.c_str());
prompt_add(prompt_inp, model_ttc, prompt_clean, false, true);
prompt_add(prompt_inp, vocab, prompt_clean, false, true);
}
prompt_add(prompt_inp, model_ttc, "<|text_end|>\n", false, true);
prompt_add(prompt_inp, vocab, "<|text_end|>\n", false, true);
// disabled to save time on tokenizing each time
// TODO: load voices from the json files
@ -549,7 +582,7 @@ it<|t_0.09|><|code_start|><|848|><|1366|><|395|><|1601|><|1513|><|593|><|1302|><
looks<|t_0.27|><|code_start|><|1281|><|1266|><|1755|><|572|><|248|><|1751|><|1257|><|695|><|1380|><|457|><|659|><|585|><|1315|><|1105|><|1776|><|736|><|24|><|736|><|654|><|1027|><|code_end|>
lovely<|t_0.56|><|code_start|><|634|><|596|><|1766|><|1556|><|1306|><|1285|><|1481|><|1721|><|1123|><|438|><|1246|><|1251|><|795|><|659|><|1381|><|1658|><|217|><|1772|><|562|><|952|><|107|><|1129|><|1112|><|467|><|550|><|1079|><|840|><|1615|><|1469|><|1380|><|168|><|917|><|836|><|1827|><|437|><|583|><|67|><|595|><|1087|><|1646|><|1493|><|1677|><|code_end|>)";
auto tmp = common_tokenize(model_ttc, voice_data, false, true);
auto tmp = common_tokenize(vocab, voice_data, false, true);
printf("\n\n");
for (int i = 0; i < tmp.size(); ++i) {
printf("%d, ", tmp[i]);
@ -715,6 +748,8 @@ lovely<|t_0.56|><|code_start|><|634|><|596|><|1766|><|1556|><|1306|><|1285|><|14
int n_past = batch.n_tokens;
int n_decode = 0;
bool next_token_uses_guide_token = true;
while (n_decode <= n_predict) {
// prepare the next batch
common_batch_clear(batch);
@ -726,7 +761,17 @@ lovely<|t_0.56|><|code_start|><|634|><|596|><|1766|><|1556|><|1306|><|1285|><|14
continue;
}
const llama_token new_token_id = common_sampler_sample(smpl[i], ctx_ttc, i_batch[i]);
llama_token new_token_id = common_sampler_sample(smpl[i], ctx_ttc, i_batch[i]);
//guide tokens help prevent hallucinations by forcing the TTS to use the correct word
if (!guide_tokens.empty() && next_token_uses_guide_token && !llama_vocab_is_control(vocab, new_token_id) && !llama_vocab_is_eog(vocab, new_token_id)) {
llama_token guide_token = guide_tokens[0];
guide_tokens.erase(guide_tokens.begin());
new_token_id = guide_token; //ensure correct word fragment is used
}
//this is the token id that always precedes a new word
next_token_uses_guide_token = (new_token_id == 198);
common_sampler_accept(smpl[i], new_token_id, true);
@ -735,9 +780,9 @@ lovely<|t_0.56|><|code_start|><|634|><|596|><|1766|><|1556|><|1306|><|1285|><|14
const auto * cands = common_sampler_get_candidates(smpl[i]);
// is it an end of generation? -> mark the stream as finished
if (llama_token_is_eog(model_ttc, new_token_id) || n_decode == n_predict) {
if (llama_vocab_is_eog(vocab, new_token_id) || n_decode == n_predict) {
std::string reason;
if (llama_token_is_eog(model_ttc, new_token_id)) {
if (llama_vocab_is_eog(vocab, new_token_id)) {
reason = "eos";
} else {
reason = "n_predict";
@ -873,7 +918,7 @@ lovely<|t_0.56|><|code_start|><|634|><|596|><|1766|><|1556|><|1306|><|1285|><|14
#if 1
// spectral operations
const int n_embd = llama_n_embd(model_cts);
const int n_embd = llama_model_n_embd(model_cts);
const float * embd = llama_get_embeddings(ctx_cts);
auto audio = embd_to_audio(embd, n_codes, n_embd, params.cpuparams.n_threads);

11
ggml/CMakeLists.txt
View File

@ -58,7 +58,8 @@ else()
set(GGML_BLAS_VENDOR_DEFAULT "Generic")
endif()
if (CMAKE_CROSSCOMPILING)
if (CMAKE_CROSSCOMPILING OR DEFINED ENV{SOURCE_DATE_EPOCH})
message(STATUS "Setting GGML_NATIVE_DEFAULT to OFF")
set(GGML_NATIVE_DEFAULT OFF)
else()
set(GGML_NATIVE_DEFAULT ON)
@ -153,6 +154,8 @@ option(GGML_CUDA_FA_ALL_QUANTS "ggml: compile all quants for FlashA
option(GGML_CUDA_GRAPHS "ggml: use CUDA graphs (llama.cpp only)" ${GGML_CUDA_GRAPHS_DEFAULT})
option(GGML_HIP "ggml: use HIP" OFF)
option(GGML_HIP_GRAPHS "ggml: use HIP graph, experimental, slow" OFF)
option(GGML_HIP_NO_VMM "ggml: do not try to use HIP VMM" ON)
option(GGML_HIP_UMA "ggml: use HIP unified memory architecture" OFF)
option(GGML_VULKAN "ggml: use Vulkan" OFF)
option(GGML_VULKAN_CHECK_RESULTS "ggml: run Vulkan op checks" OFF)
@ -185,6 +188,9 @@ option(GGML_OPENCL_PROFILING "ggml: use OpenCL profiling (increas
option(GGML_OPENCL_EMBED_KERNELS "ggml: embed kernels" ON)
option(GGML_OPENCL_USE_ADRENO_KERNELS "ggml: use optimized kernels for Adreno" ON)
# toolchain for vulkan-shaders-gen
set (GGML_VULKAN_SHADERS_GEN_TOOLCHAIN "" CACHE FILEPATH "ggml: toolchain file for vulkan-shaders-gen")
# extra artifacts
option(GGML_BUILD_TESTS "ggml: build tests" ${GGML_STANDALONE})
option(GGML_BUILD_EXAMPLES "ggml: build examples" ${GGML_STANDALONE})
@ -243,7 +249,8 @@ set(GGML_PUBLIC_HEADERS
include/ggml-metal.h
include/ggml-rpc.h
include/ggml-sycl.h
include/ggml-vulkan.h)
include/ggml-vulkan.h
include/gguf.h)
set_target_properties(ggml PROPERTIES PUBLIC_HEADER "${GGML_PUBLIC_HEADERS}")
#if (GGML_METAL)

Some files were not shown because too many files have changed in this diff Show More