A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 11:00 UTC
fix amd triton mla path (#17871)
Shiyan Deng committed
May 11, 2025 at 00:53 UTC
eea22a56ab08fb018e8fc51d1bf988cb85f37115
1 file changed
+1
-1
vllm/attention/backends/mla/common.py
+1
-1
@@ -1063,7 +1063,7 @@ class MLACommonImpl(MLAAttentionImpl[T], Generic[T]):
1063
softmax_scale,
1064
None, # bias
1065
)
1066
- if is_vllm_fa:
1066
+ elif is_vllm_fa:
1067
attn_out = self.flash_attn_varlen_func(
1068
q=q,
1069
k=k,