[Bugfix] Fix Qwen2_5_VLForConditionalGeneration packed_modules_mapping (#12905)

306923da · Jee Jee Li · GitHub · 32431583 · 306923da
Unverified Commit 306923da authored Feb 08, 2025 by Jee Jee Li Committed by GitHub Feb 07, 2025
Show whitespace changes
Inline Side-by-side

Showing with 5 additions and 2 deletions

vllm/model_executor/models/qwen2_5_vl.py vllm/model_executor/models/qwen2_5_vl.py +5 -2

No files found.
--- a/vllm/model_executor/models/qwen2_5_vl.py
+++ b/vllm/model_executor/models/qwen2_5_vl.py
@@ -760,9 +760,12 @@ class Qwen2_5_VLForConditionalGeneration(nn.Module, SupportsMultiModal,
            "q_proj",
            "k_proj",
            "v_proj",
-        ]
+        ],
+        "gate_up_proj": [
+            "gate_proj",
+            "up_proj",
+        ],
    }
-
    # LoRA specific attributes, TODO: double check
    supported_lora_modules = [
        "qkv_proj",