vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

ggml : add IQ2 to test-backend-ops + refactoring (#4990)

Browse source 
* ggml : add IQ2 to test-backend-ops + refactoring

ggml-ci

* cuda : update supports_op for IQ2

ggml-ci

* ci : enable LLAMA_CUBLAS=1 for CUDA nodes

ggml-ci

* cuda : fix out-of-bounds-access in `mul_mat_vec_q`

ggml-ci

* tests : avoid creating RNGs for each Q tensor

ggml-ci

* tests : avoid creating RNGs for each tensor

ggml-ci
This commit is contained in:
Georgi Gerganov 2024-01-17 18:54:56 +02:00 • committed by GitHub
parent ba69bbc84c
commit 38566680cd
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
9 changed files with 128 additions and 87 deletions
Download patch file Download diff file Expand all files Collapse all files

20
ggml.h
View file

@ -2065,6 +2065,18 @@ extern "C" {
// quantization
//
// - ggml_quantize_init can be called multiple times with the same type
// it will only initialize the quantization tables for the first call or after ggml_quantize_free
// automatically called by ggml_quantize_chunk for convenience
//
// - ggml_quantize_free will free any memory allocated by ggml_quantize_init
// call this at the end of the program to avoid memory leaks
//
// note: these are thread-safe
//
GGML_API void ggml_quantize_init(enum ggml_type type);
GGML_API void ggml_quantize_free(void);
// TODO: these would probably get removed in favor of the more general ggml_quantize_chunk
GGML_API size_t ggml_quantize_q4_0(const float * src, void * dst, int n, int k, int64_t * hist);
GGML_API size_t ggml_quantize_q4_1(const float * src, void * dst, int n, int k, int64_t * hist);
@ -2078,13 +2090,13 @@ extern "C" {
GGML_API size_t ggml_quantize_q5_K(const float * src, void * dst, int n, int k, int64_t * hist);
GGML_API size_t ggml_quantize_q6_K(const float * src, void * dst, int n, int k, int64_t * hist);
// some quantization type cannot be used without an importance matrix
GGML_API bool ggml_quantize_requires_imatrix(enum ggml_type type);
// calls ggml_quantize_init internally (i.e. can allocate memory)
GGML_API size_t ggml_quantize_chunk(enum ggml_type type, const float * src, void * dst,
int start, int nrows, int n_per_row, int64_t * hist, const float * imatrix);
// These are needed for IQ2_XS and IQ2_XXS quantizations
GGML_API void ggml_init_iq2_quantization(enum ggml_type type);
GGML_API void ggml_deinit_iq2_quantization(enum ggml_type type);
//
// gguf
//