Update app.py

app.py

@@ -1,9 +1,12 @@
 import gradio as gr
 from huggingface_hub import InferenceClient
 import os
+from threading import Event
 
 hf_token = os.getenv("HF_TOKEN")
 
+stop_event = Event()
+
 def get_model_response(client, messages, max_tokens, temperature, top_p):
     try:
         response = client.chat_completion(
@@ -11,13 +14,22 @@ def get_model_response(client, messages, max_tokens, temperature, top_p):
             max_tokens=max_tokens,
             temperature=temperature,
             top_p=top_p,
-            stream=False  # 스트리밍을 비활성화하여 전체 응답을 한 번에 받습니다.
+            stream=True
         )
-        return response.choices[0].message.content
+        full_response = ""
+        for message in response:
+            if stop_event.is_set():
+                break
+            token = message.choices[0].delta.content if hasattr(message.choices[0], 'delta') else message.choices[0].text
+            if token:
+                full_response += token
+                yield full_response
+        stop_event.clear()
     except Exception as e:
-        return f"모델 추론 실패: {str(e)}"
+        yield f"모델 추론 실패: {str(e)}"
 
 def respond(message, history, system_message, max_tokens, temperature, top_p, selected_model):
+    stop_event.clear()
     try:
         client = InferenceClient(model=selected_model, token=hf_token)
         
@@ -25,12 +37,17 @@ def respond(message, history, system_message, max_tokens, temperature, top_p, se
         messages.extend([{"role": "user" if i % 2 == 0 else "assistant", "content": m} for h in history for i, m in enumerate(h) if m])
         messages.append({"role": "user", "content": message})
         
-        response = get_model_response(client, messages, max_tokens, temperature, top_p)
+        response = ""
+        for partial_response in get_model_response(client, messages, max_tokens, temperature, top_p):
+            response = partial_response
+            yield "", history + [(message, response)]
         
-        history.append((message, response))
-        return "", history
     except Exception as e:
-        return "", history + [(message, f"오류 발생: {str(e)}")]
+        yield "", history + [(message, f"오류 발생: {str(e)}")]
+
+def stop_generation():
+    stop_event.set()
+    return "생성이 중단되었습니다."
 
 models = {
     "deepseek-ai/DeepSeek-Coder-V2-Instruct": "DeepSeek-Coder-V2-Instruct",
@@ -47,6 +64,7 @@ with gr.Blocks() as demo:
 
     with gr.Row():
         regenerate = gr.Button("🔄 재생성")
+        stop = gr.Button("🛑 생성 중단")
         clear = gr.Button("🗑️ 대화 내역 지우기")
 
     with gr.Accordion("추가 설정", open=True):
@@ -62,10 +80,11 @@ with gr.Blocks() as demo:
 
     send.click(respond, inputs=[msg, chatbot, system_message, max_tokens, temperature, top_p, model], outputs=[msg, chatbot])
     msg.submit(respond, inputs=[msg, chatbot, system_message, max_tokens, temperature, top_p, model], outputs=[msg, chatbot])
-    regenerate.click(lambda h: respond(h[-1][0] if h else "", h[:-1], *h.values()), inputs=[chatbot, system_message, max_tokens, temperature, top_p, model], outputs=[msg, chatbot])
+    regenerate.click(lambda h, s, m, t, p, mod: respond(h[-1][0] if h else "", h[:-1], s, m, t, p, mod), inputs=[chatbot, system_message, max_tokens, temperature, top_p, model], outputs=[msg, chatbot])
+    stop.click(stop_generation, inputs=[], outputs=[msg])
     clear.click(lambda: (None, None), outputs=[msg, chatbot])
 
 if __name__ == "__main__":
     if not hf_token:
         print("경고: HF_TOKEN 환경 변수가 설정되지 않았습니다. 일부 모델에 접근할 수 없을 수 있습니다.")
-    demo.launch()
+    demo.launch(share=True)