Use seq_len_fill_value in the cuda graph runners (#7233)

2025-06-16 15:57:07 -07:00
parent 8e2363dc15
commit c64290dcb5
7 changed files with 19 additions and 19 deletions
@@ -364,7 +364,7 @@ class FlashInferMLAAttnBackend(AttentionBackend):
            raise ValueError(f"Invalid forward mode: {forward_mode=}")

    def get_cuda_graph_seq_len_fill_value(self):
-        return 0
+        return 1

    def forward_extend(
        self,