* main : don't print special tokens with --grammar
The CLI interface was recently changed to print special control tokens
like the </s> stop message one. This token shouldn't be printed if the
grammar flag was passed, unless the grammar specifies it, because that
breaks shell-scriptability.
* main: use seperate stream for control characters
* main: use dprintf and add --ctrl-token-no-out and --ctrl-token-fd-out
* main: dprintf isn't part of the IEEE POSIX standard. Just use write().
* main: remove --ctrl-token-fd-out in favor for fcntl() based detection
* common.cpp: accidentally removed --interactive-first
* main: only merge stdout and control token if not in conversation or grammar mode
* main: rejig control token descriptor handling
* main: must check pipe status on very top of program
* main: renamed --no-special from --ctrl-token-no-out and other refactoring
* main: refactor ctrl_token_no_out --> no_special
* llama: rename llama_token_is_control_token() to llama_token_is_control()
* main: remove special token file descriptor feature (#5)
---------
Co-authored-by: Brian <mofosyne@gmail.com>
* llama : add getters for n_threads/n_threads_batch
This commit adds two new functions to the llama API. The functions
can be used to get the number of threads used for generating a single
token and the number of threads used for prompt and batch processing
(multiple tokens).
The motivation for this is that we want to be able to get the number of
threads that the a context is using. The main use case is for a
testing/verification that the number of threads is set correctly.
Signed-off-by: Daniel Bevenius <daniel.bevenius@gmail.com>
* squash! llama : add getters for n_threads/n_threads_batch
Rename the getters to llama_n_threads and llama_n_threads_batch.
Signed-off-by: Daniel Bevenius <daniel.bevenius@gmail.com>
---------
Signed-off-by: Daniel Bevenius <daniel.bevenius@gmail.com>
* Introduce bfloat16 support
Many models on Hugging Face (e.g. Mistral, TinyLLaMA) use bfloat16 as
their canonical floating point format.
┌sign
│
│ ┌exponent
│ │
│ │ ┌mantissa
│ │ │
│┌──┴───┐┌─┴───┐
0b0000000000000000 brain16
This encoding has the same number of exponent bits as float32. That
makes conversion relatively straightforward, even in the absence of
hardware support. For example, converting brain16 to binary32 means
simply shifting 16 bits to the left.
┌sign
│
│ ┌exponent
│ │
│ │ ┌mantissa
│ │ │
│┌──┴───┐┌─┴───────────────────┐
0b00000000000000000000000000000000 IEEE binary32
The issue is that converting bf16 to fp16 can result in information
loss. Only 13% of bf16 numbers can be precisely represented in fp16
which in practice ends up being 99.71% of Mistral 7b v0.2's weights
however there is currently no way other than fp32 to get the others
┌sign
│
│ ┌exponent
│ │
│ │ ┌mantissa
│ │ │
│┌─┴─┐┌─┴──────┐
0b0000000000000000 IEEE binary16
This change fixes that, by adding a bf16 data type to GGML. Support
for CPU inference has been implemented along with optimizations for
the AVX2, AVX512, and AVX512BF16 ISAs. Perplexity on Mistral 7b 0.2
improves somewhere around -0.0024 to -0.0046 compared to using fp16
* Remove GGML code that's not needed
* Minimize the GGML API surface area for BF16
* Remove bf16 luts
* Make the GGML header look nicer
* Fix documentation
* Apply ggerganov's fixes for test-backend-ops
* Add BF16 code for new ggml_validate_row_data() function
* llama : rename ctx to user_data in progress_callback
This commit renames the `ctx` parameter to `user_data` in the
`llama_progress_callback` typedef.
The motivation for this is that other callbacks use `user_data` or
`data`, and using `ctx` in this case might be confusing as it could be
confused with `llama_context`.
---------
Signed-off-by: Daniel Bevenius <daniel.bevenius@gmail.com>
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (#6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (#6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
* imatrix: save the dataset file used in the output file
* llama: support kv overrides type string string
* common: factorize KV Overrides parsing between common and server
* quantize: add imatrix n entries and dataset KV metadata
quantize: factorize KV Overrides parsing between common
#6656
* llama: remove kv override str_value initialization as it does not compile on some toolchain
* quantize: add imatrix m_last_call as `quantize.imatrix.chunks_count`
* quantize: add imatrix filename in KV
* llama: add llama_model_kv_override_free
* common: add llama_model_kv_override_free
common: free kv override if used after model loading
* llama: finally move the string KV override value to the stack
* llama : minor
* no need to add a NUL to the std::vector, std::string can be initialized from a pair of iterators.
Co-authored-by: slaren <slarengh@gmail.com>
* kv override: ensure string termination
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Co-authored-by: slaren <slarengh@gmail.com>
* add basic tensor data validation function
* add --check-tensors command line argument
tensor validation is disabled by default and can be enabled by adding
`--check-tensors` to the command line arguments.
quantize always validates tensors.
* Implement '--keep-split' to quantize model into several shards
* Add test script
* Update examples/quantize/quantize.cpp
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Split model correctly even if tensor id is out-of-order
* Update llama_model_quantize_params
* Fix preci failures
---------
Co-authored-by: z5269887 <z5269887@unsw.edu.au>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Support Llama 3 conversion
The tokenizer is BPE.
* style
* Accept suggestion
Co-authored-by: Sourab Mangrulkar <13534540+pacman100@users.noreply.github.com>
* llama : add llama_token_is_eog()
ggml-ci
* llama : auto-detect more EOT tokens when missing in KV data
* convert : replacing EOS token is a hack
* llama : fix codegemma EOT token + add TODOs
* llama : fix model type string for 8B model
---------
Co-authored-by: Sourab Mangrulkar <13534540+pacman100@users.noreply.github.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Key changes:
* BERT conversion: fix abuse of LlamaHfVocab, do not set BOS or EOS
* Nomic Embed conversion: pad vocab instead of slicing embedding tensor
* llama_tokenize: handle added special tokens like HF does
* llama_sampling_sample with default args is more naively usable
* Batches populated by either llama_batch_get_one or llama_batch_add work with default args
* Previously get_one could use the default argument
* Previously add should usually have used the last index where logits[idx] == true
* This hopefully encourages the use of llama_batch_add
* By giving expected results when using default arguments.
* Adds "negative indexing" feature to llama_get_logits_ith and llama_get_embeddings_ith
* Believed to work with any currently well behaved program
* Default arg now works for both cases (previously would give strange results for add case)
* Any non-negative number is unaffected and behaves as previously
* Negative arguments were previously invalid.
* Implemented as a special case of indexing as suggested by @compilade in https://github.com/ggerganov/llama.cpp/pull/6519
* Fixed mismatch type errors
* cited in macOS CI tests
* Missed in original updates based on PR feedback in https://github.com/ggerganov/llama.cpp/pull/6519
* llama : save and restore kv cache for single seq id
* remove trailing whitespace
* respond error in case there's no space in the kv cache
* add kv seq save restore to test case
* add --slot-save-path arg to enable save restore and restrict save location
* Returning 0 for some cases, instead of asserting.
* cleanup error cases
* rename sequence state functions
* rename state get set functions
* add previous function names back in with DEPRECATED notice
* update doc
* adjust endpoints to preferred style
* fix restoring zero cell count
* handle seq rm return value
* unused param
* keep in the size check
* fix return types
* add server test case for slot save restore
* cleanup
* add cake
* cleanup style
* add special
* removing a whole sequence never fails
* move sequence state file functionality from server to llama to match session api and add version tags
* catch exceptions on save as well
* error log messages
* check types for stricter restore
* update server doc
* readme : update API changes date
* strict filename validation
* move include, reject bom as well
* also reject empty filename
* reject whitespace and trailing dot
---------
Co-authored-by: Martin Evans <martindevans@gmail.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* iq1_m: basics
* iq1_m: basics-2
* iq1_m: CUDA dequantize works
Very 1st shot I get PPL = 9.76 for LLaMA-v2-7B.
* iq1_m: separate shifts for each group of 8 in a block
We get
PPL(LLaMA-v2-7B ) = 9.2810
PPL(LLaMA-v2-13B) = 6.8105
Not bad, but slightly higher than
sqrt(PPL(IQ1_S) * PPL(IQ2_XXS))
which is the expected outcome given that IQ1_M is
halfway between IQ1_S and IQ2_XXS in terms of bpw.
From this, we would expect
PPL = 9.14 for LLaMA-v2-7B
PPL = 6.63 for LLaMA-v2-13B
* iq1_m: go to 3-bit scales
There is slight increase in PPL, but the 0.0625 bpw reduction
in size is totally worth it.
We now have
PPL(LLaMA-v2-7B ) = 9.4469 at 1.96 bpw
PPL(LLaMA-v2-13B) = 6.8717 at 1.93 bpw
PPL(LLaMA-v2-70B) = 4.8568 at 1.85 bpw
* iq1_m: scalar dot product
* iq1_m: AVX2 dot product
* iq1_m: very slightly faster AVX2 dot product
* iq1_m: ARM_NEON dot product
Works, but very slow (10.5 t/s)
* iq1_m: Metal - dequantize works, dot product does not
* iq1_m: Metal now works
About the same performance as iq1_s.
* iq1_m: minor
* iq1_m: checking pure iq1_m quantization
It is pretty bad: PPL(LLaMA-v2-7B) = 34 if we quantize output.weight
with Q4_K.
* iiq1_m: slightly faster ARM_NEON dot product
10.5 t/s -> 11.65 t/s
* iq1_m: faster ARM_NEON dot product
11.65 t/s -> 14.9 t/s
* iq1_m: another minor ARM_NEON dot product improvement
14.9 -> 15.0 t/s
* iq1_m: small PPL improvement via super-block scale adjustment
After quantizing block scales redo the super-block scale fit.
PPL(LLaMA-v2-7B ) = 9.3346
PPL(LLaMA-v2-13B) = 6.8419
PPL(LLaMA-v2-70B) = 4.8294
PPL(Mistral-7B ) = 8.1624
* iq1_m: adapt to CUDA refactoring
* iq1_m: remove unused variable
We have progressed to warnings being errors.
* iq1_m: add to backend-ops tests
* iq1_m: fix Windows ARM
* iq1_m: use common definition of iq1m_scale_t
* cuda: assert -> NO_DEVICE_CODE
* iq1_M: PR comments
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* quantize: be able to override metadata by key
* minor : spacing
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* quantize: be able to specify the output tensor type
* quantize: be able to specify the token embedding tensor type
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* split: support in llama_model_loader
* avoid copying the entire vector
Co-authored-by: slaren <slarengh@gmail.com>
* split: move llama_tensor_offset to llama_model_loader
* llama_model_loader: PR feedbacks:
- use only one gguf_context for metadata only
- store all ggml_context in a vector as the files and mappings
- store all weights in a vector along with the source tensor
- rename ctx_gguf to meta
- rename ctx_meta to contexts
* avoid copying the entire vector
* Simplify this by making these optional, switch some layer creation tensor optional
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Handle optional tensors
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* llama_model_loader: fail if backend cannot allocate buffer
* fix mmap buffer management
* llama_model_loader: map file to backend buffer if the allocation succeeds only
* llama_model_loader: only map tensors included in the context
* llama_model_loader: minor, use same variable name for consistency, fix spacing in types cast
* llama_model_loader: fail if any of backend buffer cannot be allocated
* spacing
Co-authored-by: slaren <slarengh@gmail.com>
* fix loop over pointer
Co-authored-by: slaren <slarengh@gmail.com>
* llama_model_loader: if n_tensors declared not equals to loaded tensors in split, throw an exception instead of asserting
* llama_model_loader: ensure mappings vector has the expected size
* llama_model_loader: use at instead of operator[] if this should never add to the map.
* llama_model_loader: immediately add the backend buffer to the model buffers in order to free them if an error occurs in the next allocation. Reserve the expected size.
* llama_model_loader: be sure the model mappings has enough capacity before allocating backend buffer
* llama_model_loader: fix map -> unordered map
* llama_split_prefix: use a clearer version, not pass split path len but dest max len.
Co-authored-by: Xuan Son Nguyen <thichthat@gmail.com>
* llama : minor
ggml-ci
* llama : introduce some typedef helpers
* docs: add model shard in hot topic
* llama_model_loader: put mapping in a unique_ptr from the moment it is allocated
Co-authored-by: slaren <slarengh@gmail.com>
* fix llama_split_prefix
---------
Co-authored-by: slaren <slarengh@gmail.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Co-authored-by: Xuan Son Nguyen <thichthat@gmail.com>
* control vector api and implementation
* control-vectors : minor code style updates
* disable control vector when data == nullptr
use -1 for disabled range (also on init) in case we ever support controlling layer 0 (embeddings)
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* additional methods to read model and ctx parameters
* vocab size as a part of a model metadata
* models without vocabulary, convert.py part
* models without vocabulary, llama.cpp part
* PR clean up
* converter scrypt fixes
* llama_vocab_type update (renamed the new key)
* pr review fixes
* revert function renaming
* one more NoVocab assert
* llama : add pipeline parallelism support for batch processing with multiple CUDA GPUs
ggml-ci
* server : add -ub, --ubatch-size parameter
* fix server embedding test
* llama : fix Mamba inference for pipeline parallelism
Tested to work correctly with both `main` and `parallel` examples.
* llama : limit max batch size to n_batch
* add LLAMA_SCHED_MAX_COPIES to configure the number of input copies for pipeline parallelism
default increase to 4 (from 2)
changing this value may improve performance for some systems, but increases memory usage
* fix hip build
* fix sycl build (disable cpy_tensor_async)
* fix hip build
* llama : limit n_batch and n_ubatch to n_ctx during context creation
* llama : fix norm backend
* batched-bench : sync after decode
* swiftui : sync after decode
* ggml : allow ggml_get_rows to use multiple threads if they are available
* check n_ubatch >= n_tokens with non-casual attention
* llama : do not limit n_batch to n_ctx with non-casual attn
* server : construct batch with size of llama_n_batch
* ggml_backend_cpu_graph_compute : fix return value when alloc fails
* llama : better n_batch and n_ubatch comment
* fix merge
* small fix
* reduce default n_batch to 2048
---------
Co-authored-by: Francis Couture-Harpin <git@compilade.net>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* add gritlm example
* gritlm results match
* tabs to spaces
* comment out debug printing
* rebase to new embed
* gritlm embeddings are back babeee
* add to gitignore
* allow to toggle embedding mode
* Clean-up GritLM sample code.
* Fix types.
* Flush stdout and output ending newline if streaming.
* mostly style fixes; correct KQ_mask comment
* add causal_attn flag to llama_cparams
* gritml : minor
* llama : minor
---------
Co-authored-by: Douglas Hanley <thesecretaryofwar@gmail.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* mamba : begin working on support for Mamba SSM
* mamba : begin figuring out how to (ab)use the kv cache for Mamba
* mamba : recurrent inference almost works, but incoherent
* mamba : recurrent inference WORKS!!!
* convert : optionally use d_conv and d_state from config.json for Mamba
* mamba : refactor recurrent conv, resulting in 20% perf increase
It's still slower than I'd like, but I did not really optimize `ggml_exp` yet.
I also refactored `ggml_exp` to work with tensors with more than 2 dimensions.
* ggml : parallelize ggml_exp
This results in 8% faster token generation for Mamba-130M.
* mamba : simplify the conv step with a self-overlapping view
Turns out the conv_state can be made smaller by one column.
Note that this breaks existing GGUFs of Mamba,
because the key_value_length field is tied to the conv_state size.
Convolution with a self-overlapping view is cool!
And it's much simpler than what I initially thought would be necessary
to make the convolution step work with more than 1 token at a time.
Next step is to make the SSM step work on batches of tokens too,
and thus I need to figure out a way to make a parallel selective scan
which will keep the ssm_state small and won't make it bigger
by a factor of (n_layer * batch_size).
* llama : fix Mamba KV self size wrongly displaying as f16 instead of f32
Relatedly, I also tried to see if other types than f32 worked for the states,
but they don't, because of the operators used.
It's probably better anyway to keep lots of precision there,
since the states are small anyway.
* mamba : fix self-overlapping view depth stride
* mamba : handle batches of more than 1 token
This means running Mamba no longer crashes when using the default settings!
And probably also slightly faster prompt processing.
Both batched and non-batched processing yield the same output.
Previously, the state was not cleared when starting a sequence.
Next step is to make the KV cache API work as expected for Mamba models.
* ggml: add ggml_ssm_scan to help with parallel selective scan
If the selective scan was implemented without a custom operator,
there would be waaay too many nodes in the graph. For example,
for Mamba-130M, with a batch size of 512 (the default),
a naive selective scan could add at least 24*512=12288 nodes,
which is more than LLAMA_MAX_NODES (8192),
and that's only for the smallest Mamba model.
So it's much cleaner with a custom operator.
Not sure about the name, though.
* ggml : in ggml_ssm_scan, merge multiple rows in the same vec operation
This will help with performance on CPU if ggml_vec_mul_f32
and ggml_vec_add_f32 are ever optimized with SIMD.
* mamba : very basic quantization support
Mostly works, but there is currently no difference
between the variants of a k-quant (e.g. Q4_K_S and Q4_K_M are the same).
Most of the SSM-specific weights can be kept in f32 without affecting
the size that much, since they are relatively small.
(the linear projection weights are responsible for most of Mamba's size)
Too much quantization seems to make the state degrade quite fast, and
the model begins to output gibberish.
It seems to affect bigger models to a lesser extent than small models,
but I'm not sure by how much.
Experimentation will be needed to figure out which weights are more important
for the _M (and _L?) variants of k-quants for Mamba.
* convert : fix wrong name for layer norm weight of offical Mamba models
I was using Q-bert/Mamba-* models before, which have a slighlty different
naming scheme for the weights.
(they start with "model.layers" instead of "backbone.layers")
* mamba : fuse more steps of the SSM scan in the ggml_ssm_scan operator
This increases performance on CPU by around 30% for prompt processing,
and by around 20% for text generation.
However, it also makes the ggml_exp and ggml_soft_plus operators unused.
Whether or not they should be kept will be decided later.
* convert : for Mamba, also consider the "MambaLMHeadModel" arch name
It's the name of the class of the official implementation,
though they don't use it (yet) in the "architectures" field of config.json
* mamba : fix vocab size problems with official models
The perplexity was waaaay to high for models with a non-round vocab size.
Not sure why, but it needed to be fixed in the metadata.
Note that this breaks existing GGUF-converted Mamba models,
but **only if** the vocab size was not already rounded.
* ggml : remove ggml_exp and ggml_soft_plus
They did not exist anyway outside of this branch,
and since ggml_ssm_scan fused operations together, they are unused.
It's always possible to bring them back if needed.
* mamba : remove some useless comments
No code change.
* convert : fix flake8 linter errors
* mamba : apply suggestions from code review
* mamba : remove unecessary branch for row-wise ssm_state and C multiplication
It was previously done to avoid permuting when only one token is processed
at a time (like when generating text), but permuting is cheap,
and dynamically changing the compute graph is not future-proof.
* ggml : in ggml_ssm_scan, use more appropriate asserts
* ggml : rename the destination pointer in ggml_compute_forward_ssm_scan_f32
* mamba : multiple sequences, but one at a time
This is a step towards making this Mamba implementation usable
with the server example (the way the system prompt is kept when clearing
the client slots will need to be changed before this can work, though).
The KV cache size for this kind of model is tied to the maximum number
of sequences kept at any single time.
For now, this number is obtained from n_parallel (plus one,
to have an extra sequence to dedicate to the system prompt),
but there might be a better way to do this which won't also
make the main example use 2 cells even if only 1 is really used.
(for this specific case, --parallel 0 helps)
Simultaneous sequence processing will probably require changes to
ggml_ssm_scan, and possibly a new operator for the conv step.
* mamba : support llama_kv_cache_seq_cp
This (mis)uses the logic around K shifts, because tokens in a state
can't be shifted anyway, and because inp_K_shift has the right shape and type.
Using ggml_get_rows is a nice way to do copies, but copy chains can't work.
Fortunately, copy chains don't really seem to be used in the examples.
Each KV cell is dedicated to the sequence ID corresponding to its own index.
* mamba : use a state mask
It's cleaner than the previous heuristic of
checking for the pos of the first token in the batch.
inp_KQ_mask could not be re-used for this, because it has the wrong shape
and because it seems more suited to the next step of
simultaneous sequence processing (helping with the problem of
remembering which token belongs to which sequence(s)/state(s)).
* llama : replace the usage of n_ctx with kv_self.size in many places
* mamba : use n_tokens directly instead of n_tok
* mamba : in comments, properly refer to KV cells instead of slots
* mamba : reduce memory usage of ggml_ssm_scan
From 290.37 MiB to 140.68 MiB of CPU compute buffer size
with Mamba 3B with a batch size of 512.
The result tensor of ggml_ssm_scan was previously a big part
of the CPU compute buffer size. To make it smaller,
it does not contain the intermediate ssm states anymore.
Both y and the last ssm state are combined in the result tensor,
because it seems only a single tensor can be returned by an operator
with the way the graph is built.
* mamba : simultaneous sequence processing
A batch can now contain tokens from multiple sequences.
This is necessary for at least the parallel example, the server example,
and the HellaSwag test in the perplexity example.
However, for this to be useful, uses of llama_kv_cache_seq_rm/cp
will need to be changed to work on whole sequences.
* ggml : add ggml_ssm_conv as a new operator for the conv step of Mamba
This operator makes it possible to use and update the correct states
for each token of the batch in the same way as ggml_ssm_scan.
Other solutions which use existing operators would need loops which would
add too many nodes to the graph (at least the ones I thought of).
Using this operator further reduces the size of the CPU compute buffer
from 140.68 MiB to 103.20 MiB with Mamba 3B with a batch size of 512.
And (at least on CPU), it's a bit faster than before.
Note that "ggml_ssm_conv" is probably not the most appropriate name,
and it could be changed if a better one is found.
* llama : add inp_s_seq as a new input tensor
The most convenient implementation to select the correct state (for Mamba)
for each token is to directly get the correct index from a tensor.
This is why inp_s_seq is storing int32_t and not floats.
The other, less convenient way to select the correct state would be
to have inp_KQ_mask contain 1.0f for each state used by a token
and 0.0f otherwise. This complicates quickly fetching the first used
state of a token, and is also less efficient because a whole row
of the mask would always need to be read for each token.
Using indexes makes it easy to stop searching when there are
no more sequences for a token, and the first sequence assigned
is always very quickly available (it's the first element of each row).
* mamba : support llama_kv_cache_seq_cp copy chains
* mamba : support shifting and dividing the kv cache pos
* mamba : make the server and parallel examples work with whole sequences
A seq_id is dedicated to the system prompt in both cases.
* llama : make llama_kv_cache_seq_rm return whether it succeeded or not
* mamba : dedicate an input tensor for state copy indices
This is cleaner and makes it easier to adapt when/if token positions
(and by extension, inp_K_shift) are no longer integers.
* mamba : adapt perplexity, batched, and batched-bench examples
* perplexity : limit the max number of sequences
This adapts to what the loaded model can provide.
* llama : add llama_n_max_seq to get the upper limit for seq_ids
Used by the perplexity example.
* batched : pass n_parallel to the model's context params
This should have been there already, but it wasn't.
* batched-bench : reserve sequences to support Mamba
* batched-bench : fix tokens being put in wrong sequences
Generation quality isn't what's measured in there anyway,
but at least using the correct sequences avoids using non-consecutive
token positions.
* mamba : stop abusing attention metadata
This breaks existing converted-to-GGUF Mamba models,
but will allow supporting mixed architectures like MambaFormer
without needing to break Mamba models.
This will also allow changing the size of Mamba's states
without having to reconvert models in the future.
(e.g. using something else than d_conv - 1 columns for the conv_states
will not require breaking existing converted Mamba models again)
* gguf-py : add new KV metadata key-value pairs for Mamba
* llama : add new metadata key-value pairs for Mamba
* llama : guard against divisions by zero when n_head is 0
* mamba : rename "unlimited" KV cache property to "recurrent"
* mamba : more correctly update the "used" field of the KV cache
* ggml : in ggml_ssm_scan, use a threshold for soft_plus
This is how the official Mamba implementation does it,
and it's also what torch.nn.Softplus does.
* convert : for Mamba, fallback to internal NeoX tokenizer
The resulting models are exactly the same
as if the tokenizer.json and tokenizer_config.json of GPT-NeoX were there.
* mamba : support state saving and restoring
* ggml : implicitly pass src tensors through dst for Mamba-related ops
* mamba : clarify some comments
* server : fix cache_tokens not getting correctly resized
Otherwise, when the "we have to evaluate at least 1 token" special case
was triggered, an extra token was kept in cache_tokens even if it was
removed from the KV cache.
For Mamba, this caused useless prompt reprocessing when the previous
request triggered the above case.
* convert-hf : support new metadata keys for Mamba
For the models available at
https://huggingface.co/collections/state-spaces/transformers-compatible-mamba-65e7b40ab87e5297e45ae406
* mamba : rename metadata to be more similar to transformers library
This breaks existing converted-to-GGUF models,
but the metadata names are more "standard".
* mamba : support mamba-*-hf models
These models share their token_embd.weight with their output.weight
* mamba : add missing spaces
This is purely a formatting change.
* convert-hf : omit output.weight when identical with token_embd.weight
Only for Mamba for now, but it might be relevant for other models eventually.
Most Mamba models actually share these two tensors, albeit implicitly.
* readme : add Mamba to supported models, and add recent API changes
* mamba : move state_seq and state_mask views outside layer loop
A few tensors were also missing `struct` in front of `ggml_tensor`.
* using abort_callback from ggml to stop llama computation
* format fix
* a brief explaining comment
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Try IQ4_NL with blocks of 64 - does not look good
* iq4_xs: go to super-blocks of 256 and 6-bit scales for blocks of 32
* iq4_xs: CUDA works - 133.2 t/s
* iq4_xs: AVX2 dot product
* iq4_xs: ARM_NEON dot product
* iq4_nl: Metal implementation
As usual, Metal / Apple Silicon don't like my quants.
* iq3_xs: minor fix
* iq4_xs: shrink by using IQ3_S for attn_k and attn_q
* iq4_xs: revert using IQ3_S for attn_k and attn_v
PPL vs size is good, but CPU performance suffers: on M2 Max
TG-128 drops to 21.7 t/s from 28.8, and on a Ryzen-7950X
to 14.5 t/s from 15.8 t/s. On CUDA we have 135 t/s when
using IQ3_S vs 133 t/s with pure IQ4_XS.
* Fix CI
* iq4_xs: Added forgotten check for 256 divisibility
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* Adding IQ2_S and IQ2_M as a single cumulative commit
* Update examples/quantize/quantize.cpp
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
