mirror of
https://github.com/jart/cosmopolitan.git
synced 2025-01-31 11:37:35 +00:00
8fdb31681a
llama.com can now load weights that use the new file format which was introduced a few weeks ago. Note that, unlike llama.cpp, we will keep support for old file formats in our tool so you don't need to convert your weights when the upstream project makes breaking changes. Please note that using ggjt v3 does make avx2 inference go 5% faster for me.
22 lines
899 B
C
22 lines
899 B
C
#ifndef COSMOPOLITAN_THIRD_PARTY_GGML_V2_Q4_0_H_
|
|
#define COSMOPOLITAN_THIRD_PARTY_GGML_V2_Q4_0_H_
|
|
#if !(__ASSEMBLER__ + __LINKER__ + 0)
|
|
COSMOPOLITAN_C_START_
|
|
|
|
#define V2_QK4_0 32
|
|
typedef struct {
|
|
float d; // delta
|
|
uint8_t qs[V2_QK4_0 / 2]; // nibbles / quants
|
|
} block_v2_q4_0;
|
|
|
|
void dequantize_row_v2_q4_0(const void* restrict, float* restrict, int);
|
|
size_t ggml_quantize_v2_q4_0(const float*, void*, int, int, int64_t*);
|
|
void quantize_row_v2_q4_0(const float* restrict, void* restrict, int);
|
|
void quantize_row_v2_q4_0_reference(const float* restrict,
|
|
block_v2_q4_0* restrict, int);
|
|
void ggml_vec_dot_v2_q4_0_q8_0(const int, float* restrict, const void* restrict,
|
|
const void* restrict);
|
|
|
|
COSMOPOLITAN_C_END_
|
|
#endif /* !(__ASSEMBLER__ + __LINKER__ + 0) */
|
|
#endif /* COSMOPOLITAN_THIRD_PARTY_GGML_V2_Q4_0_H_ */
|