Update app.py

app.py

@@ -6,9 +6,16 @@ import json
 # ACCESS_TOKEN = os.getenv("HF_TOKEN")
 
 def respond(message, history, max_tokens=512, temperature=0.7, top_p=0.95):
+    # 대화 이력을 포함한 프롬프트 구성
+    if history:
+        prompt = "\n".join([f"User: {msg}" for msg, _ in history] + [f"Assistant: {res}" for _, res in history])
+        prompt += f"\nUser: {message}\nAssistant:"
+    else:
+        prompt = f"User: {message}\nAssistant:"
+
     data = {
         "model": "jinjavis:latest",
-        "prompt": message,
+        "prompt": prompt,
         "max_tokens": max_tokens,
         "temperature": temperature,
         "top_p": top_p
@@ -31,6 +38,10 @@ def respond(message, history, max_tokens=512, temperature=0.7, top_p=0.95):
                 print(f"Failed to decode JSON: {e}")
                 yield "An error occurred while processing your request."
 
+    # history 업데이트
+    history.append((message, partial_message))
+
+
 css = """
 footer {
     visibility: hidden;
@@ -48,4 +59,4 @@ demo = gr.ChatInterface(css=css,
 )
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.queue(max_size=4).launch()