Update app.py

app.py

@@ -2,11 +2,10 @@ import gradio as gr
 from dataclasses import dataclass
 import os
 from uuid import uuid4
-import json
 import requests
 
 
-DEPLOYED = os.getenv("DEPLOYED", "true").lower() != "true"
+DEPLOYED = os.getenv("DEPLOYED", "true").lower() == "true" 
 MODEL_NAME = os.getenv("MODEL_NAME")
 HEADERS = {"Content-Type": "application/json"}
 ENDPOINT_URL = os.getenv("ENDPOINT_URL")
@@ -153,7 +152,6 @@ def introduction():
             """
         )
 
-
 def chat_tab():
     def run_chat(
         message: str,
@@ -166,7 +164,6 @@ def chat_tab():
         session_id: str,
     ):
         prompt = format_chat_prompt(message, history, instructions, user_name, bot_name)
-
         payload = {
             "endpoint": MODEL_NAME,
             "data": {
@@ -182,41 +179,52 @@ def chat_tab():
                 "session_id": session_id,
             },
         }
-
         sess = requests.Session()
         full_output = ""
-        with sess.post(
-            ENDPOINT_URL, headers=HEADERS, json=payload, stream=True
-        ) as response:
-            if response.status_code == 200:
-                for chunk in response.iter_content(chunk_size=4):
-                    if chunk:
-                        decoded = chunk.decode("utf-8")
-                        full_output += decoded
-                        if full_output.endswith("User:"):
-                            yield full_output[:-5]
-                            break
-                        else:
-                            yield full_output
         last_n = 5
+        include_all_chat_history = True
         while full_output == "" and last_n > 0:
             payload["data"]["inputs"] = format_chat_prompt(
-                message, history, instructions, user_name, bot_name, False, last_n
+                message, history, instructions, user_name, bot_name, include_all_chat_history, last_n
             )
             with sess.post(
                 ENDPOINT_URL, headers=HEADERS, json=payload, stream=True
             ) as response:
-                if response.status_code == 200:
-                    for chunk in response.iter_content(chunk_size=4):
-                        if chunk:
+                if response.status_code != 200:
+                    include_all_chat_history = False             
+                    last_n -= 1
+                    continue
+                iterator = response.iter_content(chunk_size = 8)
+                try:
+                    chunk = next(iterator)
+                except (StopIteration, StopAsyncIteration):
+                    include_all_chat_history = False             
+                    last_n -= 1
+                    continue
+                while chunk:
+                    try:
+                        try:
                             decoded = chunk.decode("utf-8")
+                            chunk = b''
                             full_output += decoded
                             if full_output.endswith("User:"):
                                 yield full_output[:-5]
                                 break
                             else:
                                 yield full_output
+                                chunk = next(iterator)
+                        except UnicodeDecodeError as e:
+                            chunk += next(iterator)
+
+                    except (StopIteration, StopAsyncIteration) as si:
+                        if chunk:
+                            yield chunk.decode('UTF-8', 'ignore') 
+                        break
+            include_all_chat_history = False             
             last_n -= 1
+        
+        if full_output:
+            return "Sorry, I could not understand. Could you please rephrase your question?"
         return ""
 
     with gr.Column():