适配Qwen/Qwen2.5-VL-7B-Instruct

[gptq]

使用目前的qwen2-vl推理qwen2.5-vl报错：Traceback (most recent call last): File "/home/ubuntu/github/mindnlp_api/qwen2.5vl-test1.py", line 11, in model = Qwen2VLForConditionalGeneration.from_pretrained( File "/home/ubuntu/miniconda3/envs/mindnlp2/lib/python3.9/site-packages/mindnlp/transformers/modeling_utils.py", line 3025, in from_pretrained ) = cls._load_pretrained_model( File "/home/ubuntu/miniconda3/envs/mindnlp2/lib/python3.9/site-packages/mindnlp/transformers/modeling_utils.py", line 3426, in _load_pretrained_model raise RuntimeError(f"Error(s) in loading state_dict for {model.class.name}:\n\t{error_msg}") RuntimeError: Error(s) in loading state_dict for Qwen2VLForConditionalGeneration: size mismatch for visual.merger.mlp.2.weight: copying a param with shape (3584, 5120) from checkpoint, the shape in current model is (1280, 5120). size mismatch for visual.merger.mlp.2.bias: copying a param with shape (3584,) from checkpoint, the shape in current model is (1280,). You may consider adding ignore_mismatched_sizes=True in the model from_pretrained method.