Spaces:

natasa365
/

whisper.cpp

Running

JohannesGaessler commited on May 30

Commit

6fb9674

1 Parent(s): 1c0a5c0

CUDA: fix typo in FlashAttention code (llama/13926)

Files changed (1) hide show

ggml/src/ggml-cuda/fattn-mma-f16.cuh CHANGED Viewed

@@ -1246,7 +1246,7 @@ static __global__ void flash_attn_ext_f16(
         NO_DEVICE_CODE;
         return;
     }
-#endif __CUDA_ARCH__ == GGML_CUDA_CC_TURING
     static_assert(!mla || DKQ >= DV, "MLA needs DKQ >= DV");

         NO_DEVICE_CODE;
         return;
     }
+#endif // __CUDA_ARCH__ == GGML_CUDA_CC_TURING
     static_assert(!mla || DKQ >= DV, "MLA needs DKQ >= DV");