[MM] Pass prefix parameter to MMEncoderAttention (#33674)

Signed-off-by: shen-shanshan <467638484@qq.com>
2026-02-03 22:47:41 +08:00
parent f3d8a34671
commit 5c4f2dd6ef
15 changed files with 58 additions and 11 deletions
--- a/vllm/model_executor/models/step3_vl.py
+++ b/vllm/model_executor/models/step3_vl.py
@@ -759,7 +759,12 @@ class Step3VisionAttention(nn.Module):
        )

        # Use unified MMEncoderAttention with automatic backend selection
-        self.attn = MMEncoderAttention(self.num_heads, self.head_dim, self.scale)
+        self.attn = MMEncoderAttention(
+            self.num_heads,
+            self.head_dim,
+            self.scale,
+            prefix=prefix,
+        )

    def forward(
        self,