nikola/llama_cpp
1
0
Fork 0
Code Issues Pull Requests Actions 25 Packages Projects Releases Wiki Activity

ggml-cuda: tune RDNA3 Q6_K MMVQ nwarps (#23349)

Browse Source
This commit is contained in:
ravel7524
2026-05-20 09:52:21 +08:00
committed by GitHub
parent b28a2f372a
commit b39a7bf1b0
1 changed files with 2 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-cuda/mmvq.cu
+2
View File
@@ -359,7 +359,9 @@ static constexpr __host__ __device__ int calc_nwarps(ggml_type type, int ncols_d
case GGML_TYPE_Q5_1: case GGML_TYPE_Q5_1:
case GGML_TYPE_Q8_0: case GGML_TYPE_Q8_0:
case GGML_TYPE_Q4_K: case GGML_TYPE_Q4_K:
return 8;
case GGML_TYPE_Q6_K: case GGML_TYPE_Q6_K:
return 2;
case GGML_TYPE_IQ4_NL: case GGML_TYPE_IQ4_NL:
return 8; return 8;
default: default: