[Model] Pass param prefix to LLMHead (#24862)

Signed-off-by: whx-sjtu <2952154980@qq.com>
2025-09-17 16:01:27 +08:00
parent 03191cd8f0
commit 4a9375fe9d
58 changed files with 102 additions and 31 deletions
--- a/vllm/model_executor/models/gpt_bigcode.py
+++ b/vllm/model_executor/models/gpt_bigcode.py
@@ -302,7 +302,8 @@ class GPTBigCodeForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
            self.lm_head = ParallelLMHead(
                self.transformer.vocab_size,
                self.transformer.embed_dim,
-                org_num_embeddings=self.config.vocab_size)
+                org_num_embeddings=self.config.vocab_size,
+                prefix=maybe_prefix(prefix, "lm_head"))
        self.unpadded_vocab_size = config.vocab_size
        if lora_config:
            self.unpadded_vocab_size += lora_config.lora_extra_vocab_size