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
--- a/python/sglang/srt/layers/attention/flashinfer_backend.py
+++ b/python/sglang/srt/layers/attention/flashinfer_backend.py
@@ -440,7 +440,7 @@ class FlashInferAttnBackend(AttentionBackend):
            raise ValueError("Invalid forward mode")

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

    def forward_extend(
        self,