Update app.py

app.py

@@ -188,7 +188,7 @@ 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",
-        max_new_tokens=512*3,
+        max_new_tokens=1024,
         top_k=10,
         top_p=0.95,
         typical_p=0.95,
@@ -200,26 +200,31 @@ async def chat(query,history,sources,reports,subtype,year):
     # 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_lst.append(answer)
-    docs_html = []
-    for i, d in enumerate(context_retrieved, 1):
-        docs_html.append(make_html_source(d, i))
-    docs_html = "".join(docs_html)
-
-    previous_answer = history[-1][1]
-    previous_answer = previous_answer if previous_answer is not None else ""
-    answer_yet = previous_answer + answer_lst[0]
-    answer_yet = parse_output_llm_with_sources(answer_yet)
-    history[-1] = (query,answer_yet)
-    
-    history = [tuple(x) for x in history]
+
+    try:
+        async for op in result:
+            
+            ###-------------------------- get answers ---------------------------------------
+            answer_lst = []
+            for question, context in zip(question_lst , context_retrieved_lst):
+                answer = chain.invoke(messages)
+                answer_lst.append(answer)
+            docs_html = []
+            for i, d in enumerate(context_retrieved, 1):
+                docs_html.append(make_html_source(d, i))
+            docs_html = "".join(docs_html)
         
-    yield history,docs_html
+            previous_answer = history[-1][1]
+            previous_answer = previous_answer if previous_answer is not None else ""
+            answer_yet = previous_answer + answer_lst[0]
+            answer_yet = parse_output_llm_with_sources(answer_yet)
+            history[-1] = (query,answer_yet)
+            
+            history = [tuple(x) for x in history]
+                
+            yield history,docs_html
+    except Exception as e:
+        raise gr.Error(f"{e}")
 #process_pdf()