mirror of
https://github.com/ggml-org/llama.cpp.git
synced 2025-10-31 08:51:55 +00:00
08a0c02060
* ggml : update mul_mat_id to use the same tensor for all the experts * update cuda * minor * update metal * update test-backend-ops * fix cuda * Update ggml-metal.m Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> * update convert.py * update convert-hf-to-gguf.py * update convert.py for mixtral hf models * Update convert-hf-to-gguf.py Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> * cuda : support non-pow-2 number of experts * allow quantize to work for split and merged experts models in the same way * cleanup + disable mmap automatically with split tensors models * update imatrix * test-backend-ops : test qwen argsort * update grok model loading * llama : add merged experts tensors to the grok tensor map * minor * gguf : bump version * fix quantizing of merged experts * convert-hf-to-gguf.py : update grok (untested) * make linter happy * cuda/argsort : use shared memory instead of pool memory * convert : fix grok tensor names * metal : add support for non-pow-2 argsort * llama : more loader cleanup, better error checking * cuda : fix warning * llama : still use mmap for loading old models, but copy the data to a host buffer * add review note * llama : remove ffn tensor counting + add sanity check ggml-ci * convert : fix handling of n_experts == None ggml-ci * imatrix : fix ncall counters * llama : produce error if imatrix size does not match * quantize : terminate on errors + trace logs ggml-ci * metal : pad shared memory to 16 bytes --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
llama.cpp/examples/imatrix
Compute an importance matrix for a model and given text dataset. Can be used during quantization to enchance the quality of the quantum models. More information is available here: https://github.com/ggerganov/llama.cpp/pull/4861
Usage
./imatrix -m <some_fp_model> -f <some_training_data> [-o <output_file>] [--verbosity <verbosity_level>]
[-ofreq num_chunks] [-ow <0 or 1>] [other common params]
Here -m with a model name and -f with a file containing training data (such as e.g. wiki.train.raw) are mandatory.
The parameters in square brackets are optional and have the following meaning:
-o(or--output-file) specifies the name of the file where the computed data will be stored. If missingimatrix.datis used.--verbosityspecifies the verbosity level. If set to0, no output other than the perplexity of the processed chunks will be generated. If set to1, each time the results are saved a message is written tostderr. If>=2, a message is output each time data is collected for any tensor. Default verbosity level is1.-ofreq(or--output-frequency) specifies how often the so far computed result is saved to disk. Default is 10 (i.e., every 10 chunks)-ow(or--output-weight) specifies if data will be collected for theoutput.weighttensor. My experience is that it is better to not utilize the importance matrix when quantizingoutput.weight, so this is set tofalseby default.
For faster computation, make sure to use GPU offloading via the -ngl argument
Example
LLAMA_CUDA=1 make -j
# generate importance matrix (imatrix.dat)
./imatrix -m ggml-model-f16.gguf -f train-data.txt -ngl 99
# use the imatrix to perform a Q4_K_M quantization
./quantize --imatrix imatrix.dat ggml-model-f16.gguf ./ggml-model-q4_k_m.gguf q4_k_m