Update app.py

app.py

@@ -3,6 +3,8 @@ import requests
 import os
 import json
 from collections import deque
+import asyncio
+import aiohttp
 
 TOKEN = os.getenv("HUGGINGFACE_API_TOKEN")
 
@@ -11,7 +13,7 @@ if not TOKEN:
 
 memory = deque(maxlen=10)
 
-def respond(
+async def respond(
     message,
     history: list[tuple[str, str]],
     system_message="AI Assistant Role",
@@ -40,27 +42,27 @@ def respond(
         "temperature": temperature,
         "top_p": top_p,
         "messages": messages,
-        "stream": True  # 스트리밍 모드 활성화
+        "stream": True
     }
     
-    response = requests.post("https://api-inference.huggingface.co/v1/chat/completions", headers=headers, json=payload, stream=True)
-    
-    partial_words = ""
-    for chunk in response.iter_lines():
-        if chunk:
-            chunk_data = chunk.decode('utf-8')
-            if chunk_data.startswith("data: "):
-                chunk_data = chunk_data[6:]  # "data: " 제거
-            try:
-                response_json = json.loads(chunk_data)
-                if "choices" in response_json:
-                    delta = response_json["choices"][0].get("delta", {})
-                    if "content" in delta:
-                        content = delta["content"]
-                        partial_words += content
-                        yield partial_words
-            except json.JSONDecodeError:
-                continue
+    async with aiohttp.ClientSession() as session:
+        async with session.post("https://api-inference.huggingface.co/v1/chat/completions", headers=headers, json=payload) as response:
+            partial_words = ""
+            async for chunk in response.content:
+                if chunk:
+                    chunk_data = chunk.decode('utf-8')
+                    if chunk_data.startswith("data: "):
+                        chunk_data = chunk_data[6:]
+                    try:
+                        response_json = json.loads(chunk_data)
+                        if "choices" in response_json:
+                            delta = response_json["choices"][0].get("delta", {})
+                            if "content" in delta:
+                                content = delta["content"]
+                                partial_words += content
+                                yield partial_words
+                    except json.JSONDecodeError:
+                        continue
 
 theme = "Nymbo/Nymbo_Theme"