[Bugfix] Avoid import AttentionMetadata explicitly in Mllama (#10593)

Signed-off-by: Isotr0py <2037008807@qq.com>
2024-11-24 02:12:20 +08:00
parent 651f6c31ac
commit 04668ebe7a
5 changed files with 21 additions and 11 deletions
--- a/vllm/v1/attention/backends/flash_attn.py
+++ b/vllm/v1/attention/backends/flash_attn.py
@@ -19,7 +19,7 @@ class FlashAttentionBackend(AttentionBackend):

    @staticmethod
    def get_name() -> str:
-        return "flash-attn-vllm-v1"
+        return "FLASH_ATTN_VLLM_V1"

    @staticmethod
    def get_impl_cls() -> Type["FlashAttentionImpl"]: