diff options
| author | Kawrakow <iwankawrakow@gmail.com> | 2024-12-21 11:26:35 +0100 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2024-12-21 11:26:35 +0100 |
| commit | 93419de68f90fede135480a2717785d519df9f42 (patch) | |
| tree | 615164646770d7fdb596f04af2d887a8441f3afb /ggml/src/ggml-common.h | |
| parent | a867b919ca1e26cc828f98c35b4c6926e8e54762 (diff) | |
IQ2_S_R4 (#156)
* iq2_s_r4: Zen4
* Minor
* iq2_s_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 | 9 |
1 files changed, 9 insertions, 0 deletions
diff --git a/ggml/src/ggml-common.h b/ggml/src/ggml-common.h index 2534b461..6964f5e6 100644 --- a/ggml/src/ggml-common.h +++ b/ggml/src/ggml-common.h @@ -428,6 +428,15 @@ typedef struct { } block_iq2_s; static_assert(sizeof(block_iq2_s) == sizeof(ggml_half) + QK_K/4 + QK_K/16, "wrong iq2_s block size/padding"); +typedef struct { + ggml_half d[4]; + uint8_t qs[QK_K/2]; + uint8_t qh[QK_K/8]; + uint8_t signs[QK_K/2]; + uint8_t scales[QK_K/8]; +} block_iq2_s_r4; +static_assert(sizeof(block_iq2_s_r4) == 4*sizeof(block_iq2_s), "wrong iq2_s_r4 block size/padding"); + // (Almost) "true" 3-bit quantization. // Due to the need to use blocks as per ggml design, it ends up using // 3.0625 bpw because of the 16-bit scale for each block of 256. |