Model: Support Qwen 72B RM model. (#3772)

2025-02-21 23:38:21 +01:00
parent e310722266
commit 20b765a26e
4 changed files with 75 additions and 1 deletions
--- a/python/sglang/srt/models/qwen2.py
+++ b/python/sglang/srt/models/qwen2.py
@@ -379,6 +379,8 @@ class Qwen2ForCausalLM(nn.Module):
                continue
            if name.startswith("model.vision_tower") and name not in params_dict:
                continue
+            if name.startswith("lm_head"):
+                continue

            for param_name, weight_name, shard_id in stacked_params_mapping:
                if weight_name not in name: