Update app.py

app.py

@@ -125,10 +125,32 @@ class QwenRecapAgent:
         self.retry_delay = retry_delay
         self.device = device_map
 
-        self.tokenizer = AutoTokenizer.from_pretrained(model_path, token=hf_token)
-        model_kwargs = {"torch_dtype": torch.bfloat16, "device_map": device_map if device_map == "auto" else None}
+        # 强制使用 Fast Tokenizer
+        try:
+            self.tokenizer = AutoTokenizer.from_pretrained(
+                model_path, 
+                token=hf_token,
+                use_fast=True,  # 强制使用 fast tokenizer
+                trust_remote_code=True
+            )
+            logging.info("Successfully loaded fast tokenizer")
+        except Exception as e:
+            logging.warning(f"Fast tokenizer failed, falling back to slow: {e}")
+            self.tokenizer = AutoTokenizer.from_pretrained(
+                model_path, 
+                token=hf_token,
+                use_fast=False,
+                trust_remote_code=True
+            )
+        
+        model_kwargs = {
+            "torch_dtype": torch.bfloat16, 
+            "device_map": device_map if device_map == "auto" else None,
+            "trust_remote_code": True
+        }
         if hf_token:
             model_kwargs["token"] = hf_token
+            
         self.model = AutoModelForCausalLM.from_pretrained(model_path, **model_kwargs)
         if device_map != "auto":
              self.model.to(device_map)
@@ -180,7 +202,7 @@ Elaborate on each core requirement to create a rich description.
             model_inputs = self.tokenizer([text], return_tensors="pt").to(self.model.device)
             
             with torch.no_grad():
-                generated_ids = self.model.generate(**model_inputs, max_new_tokens=1024, temperature=0.6)
+                generated_ids = self.model.generate(**model_inputs, max_new_tokens=4096, temperature=0.6)
             
             output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist()
             full_response = self.tokenizer.decode(output_ids, skip_special_tokens=True)