Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,7 +1,6 @@
|
|
1 |
import os
|
2 |
import urllib
|
3 |
import requests
|
4 |
-
import feedparser
|
5 |
from bs4 import BeautifulSoup
|
6 |
import torch
|
7 |
import gradio as gr
|
@@ -38,7 +37,68 @@ def fetch_news(term, num_results=2):
|
|
38 |
results.append({"link": entry.link, "text": entry.title})
|
39 |
logger.debug(f"Fetched news results: {results}")
|
40 |
return results
|
41 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
42 |
# Function to format the prompt for the language model
|
43 |
def format_prompt(user_prompt, chat_history):
|
44 |
logger.debug(f"Formatting prompt with user prompt: {user_prompt} and chat history: {chat_history}")
|
@@ -72,26 +132,40 @@ def model_inference(
|
|
72 |
if web_search:
|
73 |
logger.debug("Performing news search")
|
74 |
news_results = fetch_news(user_prompt["text"])
|
75 |
-
|
76 |
-
formatted_prompt = format_prompt(f"{user_prompt['text']} [NEWS] {
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
77 |
else:
|
78 |
formatted_prompt = format_prompt(user_prompt["text"], chat_history)
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
return response
|
95 |
else:
|
96 |
return "Image input not supported in this implementation."
|
97 |
|
@@ -154,44 +228,42 @@ chatbot = gr.Chatbot(
|
|
154 |
|
155 |
# Define Gradio interface
|
156 |
def chat_interface(user_input, history, web_search, decoding_strategy, temperature, max_new_tokens, repetition_penalty, top_p):
|
157 |
-
|
158 |
-
|
159 |
-
|
|
|
160 |
response = model_inference(
|
161 |
-
user_input,
|
162 |
-
history,
|
163 |
-
web_search,
|
164 |
-
temperature,
|
165 |
-
max_new_tokens,
|
166 |
-
repetition_penalty,
|
167 |
-
top_p,
|
168 |
-
tokenizer=tokenizer # Pass tokenizer to model_inference
|
169 |
)
|
170 |
-
|
171 |
-
|
172 |
-
return
|
173 |
|
174 |
-
#
|
175 |
interface = gr.Interface(
|
176 |
fn=chat_interface,
|
177 |
inputs=[
|
178 |
-
gr.Textbox(label="User Input"),
|
179 |
-
gr.
|
180 |
-
gr.Checkbox(label="
|
181 |
decoding_strategy,
|
182 |
temperature,
|
183 |
max_new_tokens,
|
184 |
repetition_penalty,
|
185 |
top_p
|
186 |
],
|
187 |
-
outputs=
|
188 |
-
|
189 |
-
|
190 |
-
|
191 |
-
title="OpenGPT-4o-Chatty",
|
192 |
-
description="An AI assistant capable of insightful conversations and news fetching."
|
193 |
)
|
194 |
|
195 |
-
|
196 |
-
|
197 |
-
interface.launch()
|
|
|
1 |
import os
|
2 |
import urllib
|
3 |
import requests
|
|
|
4 |
from bs4 import BeautifulSoup
|
5 |
import torch
|
6 |
import gradio as gr
|
|
|
37 |
results.append({"link": entry.link, "text": entry.title})
|
38 |
logger.debug(f"Fetched news results: {results}")
|
39 |
return results
|
40 |
+
|
41 |
+
# Function to perform a Google search and return the results
|
42 |
+
def search(term, num_results=2, lang="en", timeout=5, safe="active", ssl_verify=None):
|
43 |
+
logger.debug(f"Starting search for term: {term}")
|
44 |
+
escaped_term = urllib.parse.quote_plus(term)
|
45 |
+
start = 0
|
46 |
+
all_results = []
|
47 |
+
max_chars_per_page = 8000
|
48 |
+
|
49 |
+
with requests.Session() as session:
|
50 |
+
while start < num_results:
|
51 |
+
try:
|
52 |
+
resp = session.get(
|
53 |
+
url="https://www.google.com/search",
|
54 |
+
headers={"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:109.0) Gecko/20100101 Firefox/111.0"},
|
55 |
+
params={
|
56 |
+
"q": term,
|
57 |
+
"num": num_results - start,
|
58 |
+
"hl": lang,
|
59 |
+
"start": start,
|
60 |
+
"safe": safe,
|
61 |
+
},
|
62 |
+
timeout=timeout,
|
63 |
+
verify=ssl_verify,
|
64 |
+
)
|
65 |
+
resp.raise_for_status()
|
66 |
+
soup = BeautifulSoup(resp.text, "html.parser")
|
67 |
+
result_block = soup.find_all("div", attrs={"class": "g"})
|
68 |
+
if not result_block:
|
69 |
+
start += 1
|
70 |
+
continue
|
71 |
+
for result in result_block:
|
72 |
+
link = result.find("a", href=True)
|
73 |
+
if link:
|
74 |
+
link = link["href"]
|
75 |
+
try:
|
76 |
+
webpage = session.get(link, headers={"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:109.0) Gecko/20100101 Firefox/111.0"})
|
77 |
+
webpage.raise_for_status()
|
78 |
+
visible_text = extract_text_from_webpage(webpage.text)
|
79 |
+
if len(visible_text) > max_chars_per_page:
|
80 |
+
visible_text = visible_text[:max_chars_per_page] + "..."
|
81 |
+
all_results.append({"link": link, "text": visible_text})
|
82 |
+
except requests.exceptions.RequestException as e:
|
83 |
+
logger.error(f"Error fetching or processing {link}: {e}")
|
84 |
+
all_results.append({"link": link, "text": None})
|
85 |
+
else:
|
86 |
+
all_results.append({"link": None, "text": None})
|
87 |
+
start += len(result_block)
|
88 |
+
except Exception as e:
|
89 |
+
logger.error(f"Error during search: {e}")
|
90 |
+
break
|
91 |
+
logger.debug(f"Search results: {all_results}")
|
92 |
+
return all_results
|
93 |
+
|
94 |
+
# Function to extract visible text from HTML content
|
95 |
+
def extract_text_from_webpage(html_content):
|
96 |
+
soup = BeautifulSoup(html_content, "html.parser")
|
97 |
+
for tag in soup(["script", "style", "header", "footer", "nav"]):
|
98 |
+
tag.extract()
|
99 |
+
visible_text = soup.get_text(strip=True)
|
100 |
+
return visible_text
|
101 |
+
|
102 |
# Function to format the prompt for the language model
|
103 |
def format_prompt(user_prompt, chat_history):
|
104 |
logger.debug(f"Formatting prompt with user prompt: {user_prompt} and chat history: {chat_history}")
|
|
|
132 |
if web_search:
|
133 |
logger.debug("Performing news search")
|
134 |
news_results = fetch_news(user_prompt["text"])
|
135 |
+
news2 = ' '.join([f"Link: {res['link']}\nText: {res['text']}\n\n" for res in news_results])
|
136 |
+
formatted_prompt = format_prompt(f"{user_prompt['text']} [NEWS] {news2}", chat_history)
|
137 |
+
inputs = tokenizer(formatted_prompt, return_tensors="pt").to(DEVICE)
|
138 |
+
if model:
|
139 |
+
outputs = model.generate(
|
140 |
+
**inputs,
|
141 |
+
max_new_tokens=max_new_tokens,
|
142 |
+
repetition_penalty=repetition_penalty,
|
143 |
+
do_sample=True,
|
144 |
+
temperature=temperature,
|
145 |
+
top_p=top_p
|
146 |
+
)
|
147 |
+
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
148 |
+
else:
|
149 |
+
response = "Model is not available. Please try again later."
|
150 |
+
logger.debug(f"Model response: {response}")
|
151 |
+
return response
|
152 |
else:
|
153 |
formatted_prompt = format_prompt(user_prompt["text"], chat_history)
|
154 |
+
inputs = tokenizer(formatted_prompt, return_tensors="pt").to(DEVICE)
|
155 |
+
if model:
|
156 |
+
outputs = model.generate(
|
157 |
+
**inputs,
|
158 |
+
max_new_tokens=max_new_tokens,
|
159 |
+
repetition_penalty=repetition_penalty,
|
160 |
+
do_sample=True,
|
161 |
+
temperature=temperature,
|
162 |
+
top_p=top_p
|
163 |
+
)
|
164 |
+
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
165 |
+
else:
|
166 |
+
response = "Model is not available. Please try again later."
|
167 |
+
logger.debug(f"Model response: {response}")
|
168 |
+
return response
|
|
|
169 |
else:
|
170 |
return "Image input not supported in this implementation."
|
171 |
|
|
|
228 |
|
229 |
# Define Gradio interface
|
230 |
def chat_interface(user_input, history, web_search, decoding_strategy, temperature, max_new_tokens, repetition_penalty, top_p):
|
231 |
+
# Ensure the tokenizer is accessible within the function scope
|
232 |
+
global tokenizer
|
233 |
+
|
234 |
+
# Perform model inference
|
235 |
response = model_inference(
|
236 |
+
user_prompt=user_input,
|
237 |
+
chat_history=history,
|
238 |
+
web_search=web_search,
|
239 |
+
temperature=temperature,
|
240 |
+
max_new_tokens=max_new_tokens,
|
241 |
+
repetition_penalty=repetition_penalty,
|
242 |
+
top_p=top_p,
|
243 |
+
tokenizer=tokenizer # Pass tokenizer to the model_inference function
|
244 |
)
|
245 |
+
|
246 |
+
# Return the response
|
247 |
+
return response
|
248 |
|
249 |
+
# Define the Gradio interface components
|
250 |
interface = gr.Interface(
|
251 |
fn=chat_interface,
|
252 |
inputs=[
|
253 |
+
gr.Textbox(label="User Input", placeholder="Type your message here..."),
|
254 |
+
gr.Textbox(label="Chat History", placeholder="Chat history will appear here..."),
|
255 |
+
gr.Checkbox(label="Perform Web Search", default=False),
|
256 |
decoding_strategy,
|
257 |
temperature,
|
258 |
max_new_tokens,
|
259 |
repetition_penalty,
|
260 |
top_p
|
261 |
],
|
262 |
+
outputs=gr.Textbox(label="Assistant Response"),
|
263 |
+
live=True,
|
264 |
+
layout="vertical",
|
265 |
+
theme="compact"
|
|
|
|
|
266 |
)
|
267 |
|
268 |
+
# Launch the Gradio interface
|
269 |
+
interface.launch()
|
|