Update app.py

app.py

@@ -17,8 +17,6 @@ def run_command(command, cwd=None):
         print(f"Command succeeded: {command}")
         print(result.stdout)
 
-run_command('pip install openai')
-
 # Model configuration
 MODEL_ID = "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B"
 QUANT = "Q5_K_M"
@@ -54,7 +52,7 @@ def start_llama_server(model_path):
     time.sleep(5)
     return process
 
-# GUI-specific utilities (unchanged from your original)
+# GUI-specific utilities
 def format_time(seconds_float):
     total_seconds = int(round(seconds_float))
     hours = total_seconds // 3600
@@ -86,11 +84,14 @@ details { border: 1px solid #e0e0e0 !important; border-radius: 8px !important; p
 
 client = OpenAI(base_url="http://localhost:8080/v1", api_key="no-key-required")
 
-# 修改 user() 確保 message 為字串
+# Update the user() function to use dictionary format
 def user(message, history):
     if not isinstance(message, str):
         message = str(message)
-    return "", history + [[message, None]]
+    history = history if history is not None else []
+    # Append the user message as a dict
+    history.append({"role": "user", "content": message})
+    return "", history
 
 class ParserState:
     __slots__ = ['answer', 'thought', 'in_think', 'start_time', 'last_pos', 'total_think_time']
@@ -148,13 +149,15 @@ def format_response(state, elapsed):
         )
     return collapsible, answer_part
 
+# Modified generate_response() using dictionary-format history
 def generate_response(history, temperature, top_p, max_tokens, active_gen):
-    messages = [
-        {"role": "system", "content": "You are a helpful assistant."},
-        *[{"role": "user" if i % 2 == 0 else "assistant", "content": msg or ""} 
-          for i, (user_msg, assistant_msg) in enumerate(history[:-1])],
-        {"role": "user", "content": history[-1][0]}
-    ]
+    # Guard against empty history.
+    if not history:
+        yield []
+        return
+
+    # Build messages: system message + conversation history.
+    messages = [{"role": "system", "content": "You are a helpful assistant."}] + history
     full_response = ""
     state = ParserState()
     try:
@@ -173,14 +176,24 @@ def generate_response(history, temperature, top_p, max_tokens, active_gen):
                 full_response += chunk.choices[0].delta.content
                 state, elapsed = parse_response(full_response, state)
                 collapsible, answer_part = format_response(state, elapsed)
-                history[-1][1] = "\n\n".join(collapsible + [answer_part])
+                # Update or add the assistant reply in history
+                if history and history[-1].get("role") == "assistant":
+                    history[-1]["content"] = "\n\n".join(collapsible + [answer_part])
+                else:
+                    history.append({"role": "assistant", "content": "\n\n".join(collapsible + [answer_part])})
                 yield history
         state, elapsed = parse_response(full_response, state)
         collapsible, answer_part = format_response(state, elapsed)
-        history[-1][1] = "\n\n".join(collapsible + [answer_part])
+        if history and history[-1].get("role") == "assistant":
+            history[-1]["content"] = "\n\n".join(collapsible + [answer_part])
+        else:
+            history.append({"role": "assistant", "content": "\n\n".join(collapsible + [answer_part])})
         yield history
     except Exception as e:
-        history[-1][1] = f"Error: {str(e)}"
+        if history and history[-1].get("role") == "assistant":
+            history[-1]["content"] = f"Error: {str(e)}"
+        else:
+            history.append({"role": "assistant", "content": f"Error: {str(e)}"})
         yield history
     finally:
         active_gen[0] = False
@@ -195,8 +208,8 @@ with gr.Blocks(css=CSS) as demo:
         height=500,
         show_label=False,
         render_markdown=True,
-        value=[],  # 初始值為空列表
-        type="messages"  # 指定使用 messages 格式
+        value=[],  # initial value as an empty list
+        type="messages"  # use messages format (dict with role and content)
     )
 
     with gr.Row():
@@ -259,7 +272,7 @@ if __name__ == "__main__":
     # Start llama-server
     server_process = start_llama_server(MODEL_PATH)
     try:
-        # Launch GUI
+        # Launch GUI (set share=True if you need a public link)
         demo.launch(server_name="0.0.0.0", server_port=7860)
     finally:
         # Cleanup: terminate the server process when the GUI is closed