vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Fix #4017

Browse source
This commit is contained in:
Andrew Godfrey 2023-11-15 18:17:11 -08:00
parent 34b0a08207
commit e77094faab
1 changed files with 4 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

4
ggml-cuda.cu
View file

@ -6251,6 +6251,8 @@ static int64_t get_row_rounding(ggml_type type) {
return max_compute_capability >= CC_RDNA2 ? 128 : 64; return max_compute_capability >= CC_RDNA2 ? 128 : 64;
case GGML_TYPE_F16: case GGML_TYPE_F16:
return 1; return 1;
case GGML_TYPE_F32:
return 1;
case GGML_TYPE_Q2_K: case GGML_TYPE_Q2_K:
return max_compute_capability >= CC_RDNA2 ? 128 : 32; return max_compute_capability >= CC_RDNA2 ? 128 : 32;
case GGML_TYPE_Q3_K: case GGML_TYPE_Q3_K:
@ -6273,6 +6275,8 @@ static int64_t get_row_rounding(ggml_type type) {
return 64; return 64;
case GGML_TYPE_F16: case GGML_TYPE_F16:
return 1; return 1;
case GGML_TYPE_F32:
return 1;
case GGML_TYPE_Q2_K: case GGML_TYPE_Q2_K:
case GGML_TYPE_Q3_K: case GGML_TYPE_Q3_K:
case GGML_TYPE_Q4_K: case GGML_TYPE_Q4_K: