[Bugfix] Honor --mm_encoder_attn_backend when used (#27124)

Co-authored-by: Bradley D <4551889+bradleyhd@users.noreply.github.com> Co-authored-by: Roger Wang <hey@rogerw.io>
2025-10-23 05:09:52 -07:00
parent 3a4255c7c4
commit 570c3e1cd4
6 changed files with 10 additions and 1 deletions
--- a/vllm/model_executor/models/qwen2_vl.py
+++ b/vllm/model_executor/models/qwen2_vl.py
@@ -364,6 +364,7 @@ class Qwen2VisionAttention(nn.Module):
            maybe_get_vit_flash_attn_backend(
                self.attn_backend,
                self.use_upstream_fa,
+                attn_backend_override=attn_backend_override,
            )
        )