Update app.py

app.py

@@ -6,32 +6,31 @@ from transformers import AutoTokenizer, AutoModelForCausalLM, TextIteratorStream
 
 MODEL_ID = "NoesisLab/Spartacus-1B-Instruct"
 
-# 静态加载 Tokenizer
+# 静态加载 Tokenizer (不占 GPU)
 tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, trust_remote_code=True)
 model = None
 
 @spaces.GPU
 def respond(message, history):
     global model
+    # ZeroGPU 核心逻辑：在装饰器函数内初始化并移动到 CUDA
     if model is None:
         model = AutoModelForCausalLM.from_pretrained(
             MODEL_ID,
-            dtype=torch.float16, 
+            torch_dtype=torch.float16, 
             trust_remote_code=True,
         ).to("cuda")
 
-    # 针对旧版 Gradio 结构处理：history 是 [[q1, a1], [q2, a2]]
+    # Gradio 5.x 的 history 已经是 [{'role': 'user', 'content': '...'}, ...] 格式
+    # 直接拼接到 messages 即可
     messages = [{"role": "system", "content": "You are Spartacus, a helpful assistant."}]
-    
-    if history:
-        for user_msg, assistant_msg in history:
-            messages.append({"role": "user", "content": user_msg})
-            messages.append({"role": "assistant", "content": assistant_msg})
-    
+    messages.extend(history)
     messages.append({"role": "user", "content": message})
 
     input_ids = tokenizer.apply_chat_template(
-        messages, add_generation_prompt=True, return_tensors="pt"
+        messages, 
+        add_generation_prompt=True, 
+        return_tensors="pt"
     ).to("cuda")
 
     streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
@@ -40,11 +39,12 @@ def respond(message, history):
         input_ids=input_ids,
         streamer=streamer,
         max_new_tokens=2048,
-        temperature=0.5,
-        top_p=0.9,
+        temperature=0.6,
+        top_p=0.95,
         do_sample=True,
     )
 
+    # 启动异步生成线程
     thread = Thread(target=model.generate, kwargs=generate_kwargs)
     thread.start()
 
@@ -53,11 +53,13 @@ def respond(message, history):
         response += token
         yield response
 
-# 彻底删掉 type 参数，只保留最基础的配置
+# 使用最新版的配置参数
 demo = gr.ChatInterface(
     fn=respond,
+    type="messages", # 这需要 gradio>=5.0.0
     title="Spartacus Chat",
     description="Chat with NoesisLab/Spartacus-1B-Instruct",
+    examples=["Who are you?", "Explain the concept of Noesis."],
 )
 
 if __name__ == "__main__":