Update app.py

app.py

@@ -2,6 +2,7 @@ import os
 import urllib
 import requests
 from typing import List, Dict, Union
+from threading import Thread
 import torch
 import gradio as gr
 from bs4 import BeautifulSoup
@@ -30,7 +31,7 @@ def search(term, num_results=2, lang="en", timeout=5, safe="active", ssl_verify=
     start = 0
     all_results = []
     max_chars_per_page = 8000  # Limit the number of characters from each webpage
-    
+
     with requests.Session() as session:
         while start < num_results:
             resp = session.get(
@@ -41,6 +42,8 @@ def search(term, num_results=2, lang="en", timeout=5, safe="active", ssl_verify=
                 verify=ssl_verify,
             )
             resp.raise_for_status()
+            logging.debug(f"Raw HTML response from Google: {resp.text[:500]}")  # Log the first 500 characters of the HTML
+
             soup = BeautifulSoup(resp.text, "html.parser")
             result_block = soup.find_all("div", attrs={"class": "g"})
             if not result_block:
@@ -63,6 +66,7 @@ def search(term, num_results=2, lang="en", timeout=5, safe="active", ssl_verify=
                 else:
                     all_results.append({"link": None, "text": None})
             start += len(result_block)
+    logging.debug(f"Web search results: {all_results}")
     return all_results
 
 # Format the prompt for the language model
@@ -71,10 +75,7 @@ def format_prompt(user_prompt, chat_history):
     for item in chat_history:
         if isinstance(item, tuple):
             prompt += f"[INST] {item[0]} [/INST]"
-            if len(item) > 1:
-                prompt += f" {item[1]}</s>"
-            else:
-                prompt += "</s>"
+            prompt += f" {item[1]}</s>"
         else:
             prompt += f" [Image] "
     prompt += f"[INST] {user_prompt} [/INST]"
@@ -82,21 +83,17 @@ def format_prompt(user_prompt, chat_history):
 
 # Model inference function
 def start_inference(prompt, enable_web_search):
-    logging.debug("Starting inference")
     return next(model_inference(prompt, enable_web_search))
 
 def model_inference(prompt, enable_web_search):
-    logging.debug(f"Model inference with prompt: {prompt}, enable_web_search: {enable_web_search}")
     for response in fetch_response(prompt, enable_web_search):
         yield response
 
 def fetch_response(prompt, enable_web_search):
-    logging.debug(f"Fetching response with enable_web_search: {enable_web_search}")
     if enable_web_search:
         # Perform web search and generate text based on the retrieved results
         web_results = search(prompt)
         web2 = ' '.join([f"Link: {res['link']}\nText: {res['text']}\n\n" for res in web_results])
-        logging.debug(f"Web search results: {web2}")
         client = InferenceClient("mistralai/Mistral-7B-Instruct-v0.3")
         generate_kwargs = dict(max_new_tokens=4000, do_sample=True)
         formatted_prompt = format_prompt(
@@ -107,7 +104,6 @@ def fetch_response(prompt, enable_web_search):
         for response in stream:
             if not response.token.text == "</s>":
                 output += response.token.text
-                logging.debug(f"Streaming response: {response.token.text}")
             yield output
     else:
         # Use the microsoft/Phi-3-mini-4k-instruct model for generating text based on user prompts
@@ -119,7 +115,6 @@ def fetch_response(prompt, enable_web_search):
         for response in stream:
             if not response.token.text == "</s>":
                 output += response.token.text
-                logging.debug(f"Streaming response: {response.token.text}")
             yield output
 
 # Create a chatbot interface with a Fetch button
@@ -127,7 +122,7 @@ chatbot = gr.Interface(
     fn=start_inference,
     inputs=[
         gr.Textbox(label="User Prompt", placeholder="Enter your prompt here..."),
-        gr.Checkbox(label="Enable Web Search", value=False, info="Search Web")
+        gr.Checkbox(label="Enable Web Search", value=False, description="Search Web")
     ],
     outputs=gr.Textbox(label="Response", placeholder="Responses will appear here..."),
     live=True