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

Add quantize-stats command for testing quantization (#728)

Browse source 
Command that calculates some statistics over the errors introduced by
quantization, like mean square error, max error and some percentile errors for layer
weights. Should be useful for testing quantization improvements.

Exposes some internal state from ggml and llama for testing
This commit is contained in:
unbounded 2023-04-08 00:09:18 +02:00 • committed by GitHub
parent 698f7b5d63
commit 62cfc54f77
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
9 changed files with 415 additions and 17 deletions
Download patch file Download diff file Expand all files Collapse all files

24
ggml.h
View file

@ -783,6 +783,30 @@ int ggml_cpu_has_blas(void);
int ggml_cpu_has_sse3(void);
int ggml_cpu_has_vsx(void);
//
// Internal types and functions exposed for tests and benchmarks
//
#ifdef __cplusplus
// restrict not standard in C++
#define GGML_RESTRICT
#else
#define GGML_RESTRICT restrict
#endif
typedef void (*dequantize_row_q_t)(const void * GGML_RESTRICT x, float * GGML_RESTRICT y, int k);
typedef void (*quantize_row_q_t)(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int k);
typedef void (*vec_dot_q_t)(const int n, float * GGML_RESTRICT s, const void * GGML_RESTRICT x, const void * GGML_RESTRICT y);
typedef struct {
dequantize_row_q_t dequantize_row_q;
quantize_row_q_t quantize_row_q;
quantize_row_q_t quantize_row_q_reference;
vec_dot_q_t vec_dot_q;
} quantize_fns_t;
quantize_fns_t ggml_internal_get_quantize_fn(size_t i);
#ifdef __cplusplus
}
#endif