diff options
| author | Kawrakow <iwankawrakow@gmail.com> | 2024-10-13 13:34:30 +0300 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2024-10-13 13:34:30 +0300 |
| commit | 910a13409463f7aedb0a92be013a1b9bb50f4859 (patch) | |
| tree | 16e13e1fd3010549877408a0a62706b2bc5d5f0c /ggml/src/ggml-common.h | |
| parent | c15de3654e0002537c8052fd6d52d879e778e88c (diff) | |
IQ2_KS: 2.1875 bpw non-linear quantization (#85)
* Experimenting
* iq2k: Try make_qx_quants for the scale
Slightly better for LLaMA-3.1, Gemma-2, slightly worse for
Qwen2.5
* iq2k with make_qx_quants: adjust scale
* iq2ks: basics
* iq2_ks: CUDA works
* iq2_ks: WIP
* iq2_ks: WIP
* iq2_ks: Zen4
* iq2_ks: AVX2
* iq2_ks: scalar dot product
* iq2_ks: ARM_NEON
* iq2_ks: Metal
* iq2_ks: faster Metal
LLaMA-3.1-8B:
PP-512 = 475.22 ± 0.37 t/s
TG-128 = 45.32 ± 0.03 t/s
---------
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 7eaf7437..3a7b8989 100644 --- a/ggml/src/ggml-common.h +++ b/ggml/src/ggml-common.h @@ -456,6 +456,13 @@ typedef struct { static_assert(sizeof(block_iq2_k) == sizeof(ggml_half) + sizeof(uint16_t) + QK_K/32 + QK_K/4, "wrong iq2_k block size/padding"); typedef struct { + uint16_t extra; + uint8_t scales[QK_K/64]; + uint8_t qs[QK_K/4]; +} block_iq2_ks; +static_assert(sizeof(block_iq2_ks) == sizeof(uint16_t) + QK_K/64 + QK_K/4, "wrong iq2_ks block size/padding"); + +typedef struct { ggml_half d; uint16_t extra; uint16_t scales_h; |