diff options
| author | Kawrakow <iwankawrakow@gmail.com> | 2024-12-20 12:02:42 +0100 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2024-12-20 12:02:42 +0100 |
| commit | 4b53bc876eda6d89890711998b3ab5bf033b2199 (patch) | |
| tree | 4624c21b636cb6c5121ae911c154ff8ed5702791 /ggml/src/ggml-common.h | |
| parent | 7254f6d34024b708f15d3e6ce6926a808a4d0cc4 (diff) | |
IQ2_XXS_R4 (#154)
* iq2_xxs_r4: Zen4
Disapointing gain: 134.7 t/s -> 151.1 t/s for PP-512
TG-128 is better: 3.45 -> 4.61 t/s @ 1 thread
* Minor
* iq2_xxs_r4: NEON
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
Diffstat (limited to 'ggml/src/ggml-common.h')
| -rw-r--r-- | ggml/src/ggml-common.h | 7 |
1 files changed, 7 insertions, 0 deletions
diff --git a/ggml/src/ggml-common.h b/ggml/src/ggml-common.h index b9da9771..ad21dd50 100644 --- a/ggml/src/ggml-common.h +++ b/ggml/src/ggml-common.h @@ -397,6 +397,13 @@ typedef struct { } block_iq2_xxs; static_assert(sizeof(block_iq2_xxs) == sizeof(ggml_half) + QK_K/8*sizeof(uint16_t), "wrong iq2_xxs block size/padding"); +typedef struct { + ggml_half d[4]; + uint8_t sas[QK_K/2]; + uint8_t qs[QK_K/2]; +} block_iq2_xxs_r4; +static_assert(sizeof(block_iq2_xxs_r4) == 4*sizeof(block_iq2_xxs), "wrong iq2_xxs_r4 block size/padding"); + // 2.3125 bpw quants typedef struct { ggml_half d; |