[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/glm_ocr.py
+++ b/vllm/model_executor/models/glm_ocr.py
@@ -137,6 +137,7 @@ class GlmOcrVisionAttention(nn.Module):
            num_heads=self.num_attention_heads_per_partition,
            head_size=self.hidden_size_per_attention_head,
            scale=self.hidden_size_per_attention_head**-0.5,
+            prefix=f"{prefix}.attn",
        )
        self.apply_rotary_emb = ApplyRotaryEmb(enforce_enable=True)