diff options
| author | Iwan Kawrakow <iwan.kawrakow@gmail.com> | 2024-07-30 16:11:25 +0300 |
|---|---|---|
| committer | Kawrakow <48489457+ikawrakow@users.noreply.github.com> | 2024-08-01 09:38:06 +0200 |
| commit | 4f237d44f6d75afbb5cef39d4d6b0b35b2a517c7 (patch) | |
| tree | 097eb7ee5be6ec08d3958f747775f9225db3c7a4 /ggml/src/ggml-cuda/common.cuh | |
| parent | 36204c4ec7815f1b0f8f38c8e1cd3e6b3fef2eaa (diff) | |
iq3_k: Basics
Quantize/dequantize, CUDA dequantize.
PPL of LLaMA-3.1-8B is better than iq3_s and iq3_m.
Diffstat (limited to 'ggml/src/ggml-cuda/common.cuh')
| -rw-r--r-- | ggml/src/ggml-cuda/common.cuh | 7 |
1 files changed, 7 insertions, 0 deletions
diff --git a/ggml/src/ggml-cuda/common.cuh b/ggml/src/ggml-cuda/common.cuh index 516e74d8..fbc52aa9 100644 --- a/ggml/src/ggml-cuda/common.cuh +++ b/ggml/src/ggml-cuda/common.cuh @@ -677,6 +677,13 @@ struct ggml_cuda_type_traits<GGML_TYPE_IQ2_K> { }; template<> +struct ggml_cuda_type_traits<GGML_TYPE_IQ3_K> { + static constexpr int qk = QK_K; + static constexpr int qr = QR4_XS; + static constexpr int qi = QI4_XS; +}; + +template<> struct ggml_cuda_type_traits<GGML_TYPE_IQ4_K> { static constexpr int qk = QK_K; static constexpr int qr = QR4_XS; |