Update app.py

app.py

@@ -188,16 +188,15 @@ async def chat(query,history,sources,reports,subtype,year):
     #callbacks = [StreamingStdOutCallbackHandler()]
     llm_qa = HuggingFaceEndpoint(
         endpoint_url="https://howaqfw0lpap12sg.us-east-1.aws.endpoints.huggingface.cloud",
-        task="text-generation",
-        huggingfacehub_api_token=HF_token)
+        max_new_tokens=1024,
+        huggingfacehub_api_token=HF_token,)
 
     # create rag chain
     chat_model = ChatHuggingFace(llm=llm_qa)
-    chain = chat_model | StrOutputParser()
     ###-------------------------- get answers ---------------------------------------
     answer_lst = []
     for question, context in zip(question_lst , context_retrieved_lst):
-        answer = chain.invoke(messages)
+        answer = chat_model.invoke(messages)
         answer_lst.append(answer)
     docs_html = []
     for i, d in enumerate(context_retrieved, 1):