Skip to content

Commit 00c146a

Browse files
committed
Update fattn.cu
1 parent 8de69c9 commit 00c146a

1 file changed

Lines changed: 4 additions & 0 deletions

File tree

ggml/src/ggml-cuda/fattn.cu

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -465,6 +465,8 @@ static best_fattn_kernel ggml_cuda_get_best_fattn_kernel(const int device, const
465465
case GGML_TYPE_Q4_1:
466466
case GGML_TYPE_Q5_0:
467467
case GGML_TYPE_Q5_1:
468+
case GGML_TYPE_Q6_0:
469+
case GGML_TYPE_IQ4_NL:
468470
#ifndef GGML_CUDA_FA_ALL_QUANTS
469471
return BEST_FATTN_KERNEL_NONE;
470472
#endif // GGML_CUDA_FA_ALL_QUANTS
@@ -490,6 +492,8 @@ static best_fattn_kernel ggml_cuda_get_best_fattn_kernel(const int device, const
490492
case GGML_TYPE_Q4_1:
491493
case GGML_TYPE_Q5_0:
492494
case GGML_TYPE_Q5_1:
495+
case GGML_TYPE_Q6_0:
496+
case GGML_TYPE_IQ4_NL:
493497
case GGML_TYPE_Q4_0:
494498
case GGML_TYPE_Q8_0:
495499
if (K->ne[0] != 256 && K->ne[0] != 128) {

0 commit comments

Comments
 (0)