Update app.py

app.py

@@ -6,7 +6,7 @@ from transformers import AutoTokenizer, AutoModelForCausalLM, TextIteratorStream
 
 MODEL_ID = "NoesisLab/Spartacus-1B-Instruct"
 
-# Tokenizer 静态加载
+# 静态加载 Tokenizer
 tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, trust_remote_code=True)
 model = None
 
@@ -16,15 +16,21 @@ def respond(message, history):
     if model is None:
         model = AutoModelForCausalLM.from_pretrained(
             MODEL_ID,
-            torch_dtype=torch.float16,
+            dtype=torch.float16, # 修复弃用警告：torch_dtype -> dtype
             trust_remote_code=True,
         ).to("cuda")
 
-    # 适配旧版 Gradio 的 history 格式: [[user, assistant], [user, assistant]]
+    # 更加鲁棒的 history 处理逻辑
     messages = [{"role": "system", "content": "You are Spartacus, a helpful assistant."}]
-    for user_msg, assistant_msg in history:
-        messages.append({"role": "user", "content": user_msg})
-        messages.append({"role": "assistant", "content": assistant_msg})
+    
+    for entry in history:
+        # 自动兼容 字典格式 {'role': '...', 'content': '...'}
+        if isinstance(entry, dict):
+            messages.append(entry)
+        # 自动兼容 元组/列表格式 [user, assistant]
+        elif isinstance(entry, (list, tuple)) and len(entry) == 2:
+            messages.append({"role": "user", "content": entry[0]})
+            messages.append({"role": "assistant", "content": entry[1]})
     
     messages.append({"role": "user", "content": message})
 
@@ -49,11 +55,12 @@ def respond(message, history):
     response = ""
     for token in streamer:
         response += token
-        yield response # 旧版 ChatInterface 需要 yield 完整的累计字符串
+        yield response
 
-# 移除 type="messages" 参数
+# 强制指定 type="tuples" 来适配上面的 history 处理，避免 unpacked 错误
 demo = gr.ChatInterface(
     fn=respond,
+    type="tuples", # 显式声明，确保 entry 是 [user, bot] 结构
     title="Spartacus Chat",
     description="Chat with NoesisLab/Spartacus-1B-Instruct",
 )