Upload modeling_qwen2_5_vl.py

modeling_qwen2_5_vl.py

@@ -1,9 +1,22 @@
-from transformers import AutoModelForCausalLM
+from transformers import PreTrainedModel
+from transformers import AutoModelForCausalLM  # 用于模型内部实际加载
+from configuration_qwen2_5_vl import Qwen2_5_VLConfig  # 你自定义的配置类
 import torch
 
-class QWenVLChatModel(AutoModelForCausalLM):
+class QWenVLChatModel(PreTrainedModel):
+    config_class = Qwen2_5_VLConfig
+    base_model_prefix = "qwen2_5_vl"
+
+    def __init__(self, config):
+        super().__init__(config)
+        # 用 AutoModelForCausalLM 来加载具体模型架构
+        self.model = AutoModelForCausalLM.from_config(config)
+
+    def forward(self, *args, **kwargs):
+        return self.model(*args, **kwargs)
+
     def chat(self, tokenizer, query: str, image=None, history=None, **kwargs):
         inputs = tokenizer(query, return_tensors="pt").to(self.device)
         with torch.no_grad():
-            outputs = self.generate(**inputs, max_new_tokens=512)
+            outputs = self.model.generate(**inputs, max_new_tokens=512)
         return tokenizer.decode(outputs[0], skip_special_tokens=True), history