diff options
| author | Kawrakow <iwankawrakow@gmail.com> | 2025-06-11 15:01:34 +0300 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2025-06-11 15:01:34 +0300 |
| commit | 3f54b4978672cc236bf38e707b24f036cb37f3f7 (patch) | |
| tree | 41221f8bfc21f1fa4689df962d98aff95fc2df68 /ggml/src/ggml.c | |
| parent | 69af3f5990327df4f8dcb05817825802b7e6bed8 (diff) | |
Faster iq1_s GEMM via repacking to Q8_0_R8 (#517)
TG is slightly faster too - 24.4 vs 23.1 t/s on the
Ryzen-5975WX
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
Diffstat (limited to 'ggml/src/ggml.c')
| -rw-r--r-- | ggml/src/ggml.c | 4 |
1 files changed, 4 insertions, 0 deletions
diff --git a/ggml/src/ggml.c b/ggml/src/ggml.c index bf5a8519..e4e954a1 100644 --- a/ggml/src/ggml.c +++ b/ggml/src/ggml.c @@ -1205,7 +1205,11 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = { .from_float = quantize_row_iq1_s, .from_float_ref = (ggml_from_float_t)quantize_row_iq1_s_ref, .vec_dot = ggml_vec_dot_iq1_s_q8_K, +#ifdef __AVX2__ + .vec_dot_type = GGML_TYPE_Q8_2_X4, +#else .vec_dot_type = GGML_TYPE_Q8_K, +#endif .nrows = 1, .row_meta_size = 0, }, |