Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-28 15:18:26 +01:00
Code Issues Packages Projects Releases Wiki Activity

iq3_s_mult: another alternative multiplier

Browse Source
This commit is contained in:
Iwan Kawrakow 2024-03-03 13:13:52 +02:00
parent 8b713a987e
commit dbe98dfe70
2 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
ggml-cuda.cu
View File

@ -2374,8 +2374,8 @@ static __global__ void dequantize_block_iq3_xxs(const void * __restrict__ vx, ds
// Better (lower PPL), but requires more bit twidling, so slower
#define IQ3S_MULTIPLIER 190842953LL
#else
//#define IQ3S_MULTIPLIER 898886
#define IQ3S_MULTIPLIER 842866
#define IQ3S_MULTIPLIER 898886
//#define IQ3S_MULTIPLIER 842866
#endif
template<typename dst_t>

4
ggml-quants.c
View File

@ -4125,8 +4125,8 @@ void dequantize_row_iq3_xxs(const block_iq3_xxs * restrict x, float * restrict y
// Best PPL
#define IQ3S_MULTIPLIER 190842953
#else
//#define IQ3S_MULTIPLIER 898886
#define IQ3S_MULTIPLIER 842866
#define IQ3S_MULTIPLIER 898886
//#define IQ3S_MULTIPLIER 842866
#endif
#define IQ3S_BITS 3