[MM] Align the prefix of MMEncoderAttention with Attention (#33750)

Signed-off-by: shen-shanshan <467638484@qq.com>
2026-02-04 12:07:30 +08:00
parent 4dffc5e044
commit 9fb27dd3b3
17 changed files with 17 additions and 15 deletions
--- a/vllm/model_executor/models/mllama4.py
+++ b/vllm/model_executor/models/mllama4.py
@@ -257,7 +257,7 @@ class Llama4VisionAttention(nn.Module):
            self.num_local_heads,
            self.head_dim,
            self.scaling,
-            prefix=prefix,
+            prefix=f"{prefix}.attn",
        )

        if use_data_parallel: