Update app.py

app.py

@@ -60,6 +60,8 @@ def qwen_api(user_message, top_p=0.9,temperature=0.7, system_message='', max_tok
         yield response
     return response
 
+os.environ["OPENAI_API_BASE"] = "https://api-inference.huggingface.co/v1/"
+os.environ["OPENAI_API_KEY"] = TOKEN
 llm = ChatOpenAI(
         model="meta-llama/Meta-Llama-3-8B-Instruct",
         temperature=0.8,)