* Remove split metadata when quantize model shards
* Find metadata key by enum
* Correct loop range for gguf_remove_key and code format
* Free kv memory
---------
Co-authored-by: z5269887 <z5269887@unsw.edu.au>
* cmake : make WHISPER_NO_AVX512=ON disable all subsets of AVX-512
Previously it happened only for MSVC, but it makes sense to have the
same behavior for other compilers too.
* make : reorder x86 ISA extensions in chronological order
And update compiler flags at the end to ease modifying conditions.
* make : support WHISPER_NO_AVX512=1 for disabling all AVX-512 subsets.
That way you do not have to override each AVX-512 subset setting
individually if it has been turned on during autodetection.
fft_out needs to be twice the frame_size, not the frame_step. It is resized in fft() anyway, but this change prevents an unnecessary reallocation.
n_fft must match the mel filter size, so it is best not to calculate it from the framesize.
We only need to get the magnitudes for half the spectrum since the other half is a mirror and not used in the mel filter loop later.
* make : add AVX512 detection to Makefile and CMakeLists.txt
* make : autodetect more AVX512 instruction subsets
* cmake : do not default to AVX512, must be enabled explicitly
* cmake : enable a set of AVX512 subsets, when AVX512 is turned on
* make : consolidate AVX512 subsets, add AVX512 VBMI
* cmake : revert to NO AVX512 setting, add settings for AVX512 VNNI and VBMI
* make : re-introduce AVX512VNNI back
* cmake : remove superfluous comment line
* Add Command R Plus GGUF
* Add Command R Plus GGUF
* Loading works up to LayerNorm2D
* Export new tensors in 1D so they are not quantized.
* Fix embedding layer based on Noeda's example
* Whitespace
* Add line
* Fix unexpected tokens on MPS. Re-add F16 fix. ((Noeda)
* dranger003: Fix block index overflow in CUDA dequantizing.
* Reverted blocked multiplication code as it still has issues and could affect other Llama arches
* export norms as f32
* fix overflow issues during quant and other cleanup
* Type convention
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* dranger003: Fix more int overflow during quant.
---------
Co-authored-by: S <seast@Ss-Mac-Studio.local>
Co-authored-by: S <s@example.com>
Co-authored-by: slaren <slarengh@gmail.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* The "main" example now allows a response-file as the sole parameter.
A response-file is a text file with command-line parameters, one per line.
Prefix the name of the response-file with "@" to identify it as such.
It's used under MS Windows to work around command-line length limits.
It may be useful under other platforms to simplify character-escaping.
* minor : style
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Allow a regular expression to describe tokens to suppress.
Example: --suppress-tokens-re "[,\.]|[ ]?[0-9]+" will suppress commas, periods, and numeric tokens.
Technique inspired by https://github.com/openai/whisper/discussions/1041
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Blind change to fix Java test.
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* 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>
* Fix Vulkan no kv offload incoherence
* Add k-quant mul mat mat shaders
* Rework working buffer allocation, reduces vram use noticeably
Clean up cpu assist code, replaced with ggml-backend offload function
* Default to all dedicated GPUs
* Add fallback for integrated GPUs if no dedicated GPUs are found
* Add debug info which device is allocating memory
* Fix Intel dequant issue
Fix validation issue
* Fix Vulkan GGML_OP_GET_ROWS implementation
* Clean up merge artifacts
* Remove Vulkan warning
