Update app.py

app.py

@@ -105,9 +105,21 @@ def generate_completion(prompt, max_tokens=256, temperature=0.7):
         # 构建提示词 - 使用Qwen模型的对话格式
         text = f"<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant\n"
         
-        # 编码输入
-        inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=1024)
-        inputs = inputs.to(DEVICE)
+        # 记录文本类型和长度
+        logger.info(f"输入文本类型: {type(text)}, 长度: {len(text)}")
+        
+        # 编码输入 - 使用更安全的编码方式
+        encoding = tokenizer.encode_plus(
+            text,
+            return_tensors="pt",
+            truncation=True,
+            max_length=1024,
+            padding="max_length" if tokenizer.pad_token_id is not None else False
+        )
+        inputs = {
+            "input_ids": encoding["input_ids"].to(DEVICE),
+            "attention_mask": encoding["attention_mask"].to(DEVICE)
+        }
         
         # 生成响应
         with torch.no_grad():
@@ -121,7 +133,7 @@ def generate_completion(prompt, max_tokens=256, temperature=0.7):
             )
         
         # 解码响应
-        response = tokenizer.decode(outputs[0][len(inputs[0]):], skip_special_tokens=True)
+        response = tokenizer.decode(outputs[0][len(inputs["input_ids"][0]):], skip_special_tokens=True)
         
         # 清理特殊标记
         response = response.replace("<|im_end|>", "").strip()
@@ -264,10 +276,22 @@ async def create_chat_completion(
         max_tokens = data.get("max_tokens", MAX_TOKENS)
         temperature = data.get("temperature", 0.7)
         
+        # 检查消息格式
+        if not messages or not isinstance(messages, list):
+            return JSONResponse(
+                status_code=400,
+                content={
+                    "error": {
+                        "message": "无效的消息格式",
+                        "type": "invalid_request_error"
+                    }
+                }
+            )
+        
         # 从消息中提取用户提示
         user_message = ""
         for msg in messages:
-            if msg.get("role") == "user":
+            if isinstance(msg, dict) and msg.get("role") == "user":
                 user_message = msg.get("content", "")
                 break
         
@@ -343,10 +367,19 @@ async def openclaw_chat_api(
         data = await request.json()
         messages = data.get("messages", [])
         
+        # 检查消息格式
+        if not messages or not isinstance(messages, list):
+            return JSONResponse(
+                status_code=400,
+                content={
+                    "error": "无效的消息格式"
+                }
+            )
+        
         # 提取用户消息
         user_message = ""
         for msg in messages:
-            if msg.get("role") == "user":
+            if isinstance(msg, dict) and msg.get("role") == "user":
                 user_message = msg.get("content", "")
                 break