ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-02-24 15:14:10 +00:00
Code Issues Packages Projects Releases Wiki Activity

WIP - working basic iq2_kt

Browse Source
This commit is contained in:
Andrew Keen Chan
2025-05-18 21:07:22 +00:00
parent 3cc0de96a6
commit 7561158313
1 changed files with 13 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
ggml/src/iqk/iqk_mul_mat.cpp
View File

@@ -3033,9 +3033,9 @@ static inline uint32_t trellis_next(uint32_t& val) {
}
static inline float trellis_gen(uint32_t& val, uint32_t* s) {
const ggml_half * h = (const ggml_half *)s;
const ggml_fp16_t * h = (const ggml_fp16_t *)s;
s[0] = trellis_next(val);
return (float)(h[0] + h[1]);
return GGML_FP16_TO_FP32(h[0]) + GGML_FP16_TO_FP32(h[1]);
}
template <int nrc_y>
@@ -3057,11 +3057,19 @@ static void mul_mat_q2_KT_q8_K_T(int n, const void * vx, size_t bx, const DataIn
for (int i = 0; i < nb; ++i) {
const uint16_t * ql = (const uint16_t *)x[i].ql;
uint32_t val = ql[0] + 4096;
uint32_t val;
for (int j = 0; j < QK_K; ++j) {
const float x_scale = iq4k_values[x[i].scales[j/4] & 0xf];
if (j % 8 == 0) { val = ql[j/8] + 4096; }
float x_scale;
if (j < 128) {
x_scale = iq4k_values[x[i].scales[j/32 % 4] & 0xf];
} else {
x_scale = iq4k_values[x[i].scales[j/32 % 4] >> 4];
}
float x_val = trellis_gen(val, s);
x_val *= x_scale;
for (int iy = 0; iy < nrc_y; ++iy) {
accd[iy] += (y[iy]->d*y[iy]->qs[j]) * x_scale * trellis_gen(val, s);
accd[iy] += (y[iy][i].d*y[iy][i].qs[j]) * x_val;
}
}
}