[Misc][Model][Refactor] Pass the prefix into Linear layers (#28259)

Signed-off-by: MengqingCao <cmq0113@163.com>
2025-11-07 19:38:38 +08:00
parent 7bdb42b2f2
commit 1958bda9b4
26 changed files with 190 additions and 25 deletions
--- a/vllm/model_executor/models/qwen.py
+++ b/vllm/model_executor/models/qwen.py
@@ -102,12 +102,14 @@ class QWenAttention(nn.Module):
            self.total_num_heads,
            bias=True,
            quant_config=quant_config,
+            prefix=f"{prefix}.c_attn",
        )
        self.c_proj = RowParallelLinear(
            self.total_num_heads * self.head_dim,
            hidden_size,
            bias=False,
            quant_config=quant_config,
+            prefix=f"{prefix}.c_proj",
        )
        self.scaling = self.head_dim**-0.5