Spaces:
Runtime error
Runtime error
Create app.py
Browse files
app.py
ADDED
|
@@ -0,0 +1,247 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import gradio as gr
|
| 2 |
+
from youtube_transcript_api import YouTubeTranscriptApi
|
| 3 |
+
from openai import OpenAI
|
| 4 |
+
import os
|
| 5 |
+
import yt_dlp
|
| 6 |
+
import logging
|
| 7 |
+
from langgraph.graph import StateGraph, END
|
| 8 |
+
from typing import Dict, List, TypedDict, Optional
|
| 9 |
+
from langgraph.checkpoint.memory import MemorySaver
|
| 10 |
+
|
| 11 |
+
# Set up logging to debug issues
|
| 12 |
+
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
|
| 13 |
+
logger = logging.getLogger(__name__)
|
| 14 |
+
|
| 15 |
+
# Initialize OpenAI client
|
| 16 |
+
api_key = os.getenv("OPENAI_API_KEY")
|
| 17 |
+
if not api_key:
|
| 18 |
+
raise ValueError("OPENAI_API_KEY environment variable is not set. Please set it in Hugging Face Secrets or your environment.")
|
| 19 |
+
client = OpenAI(api_key=api_key)
|
| 20 |
+
|
| 21 |
+
# Define the state for LangGraph
|
| 22 |
+
class SummarizerState(TypedDict):
|
| 23 |
+
urls: List[str]
|
| 24 |
+
titles: List[str]
|
| 25 |
+
transcripts: List[str]
|
| 26 |
+
summaries: List[str]
|
| 27 |
+
master_summary: str
|
| 28 |
+
errors: List[str]
|
| 29 |
+
master_output: str
|
| 30 |
+
individual_output: str
|
| 31 |
+
error_output: str
|
| 32 |
+
|
| 33 |
+
# Instructions for the front-end
|
| 34 |
+
INSTRUCTIONS = """
|
| 35 |
+
**YouTube Video Summarizer**
|
| 36 |
+
|
| 37 |
+
- **Left Panel**: Enter a YouTube URL and click "Add URL" to include it (up to 3 videos). View added URLs and their titles below. Click "Clear" to start over.
|
| 38 |
+
- **Right Panel**: Click "Summarize" to generate a master summary (combining insights from all videos) and individual summaries for each video in bullet points.
|
| 39 |
+
"""
|
| 40 |
+
|
| 41 |
+
# Function to get video title using yt-dlp
|
| 42 |
+
def get_video_title(video_url: str) -> str:
|
| 43 |
+
try:
|
| 44 |
+
ydl_opts = {
|
| 45 |
+
'quiet': True,
|
| 46 |
+
'no_warnings': True,
|
| 47 |
+
}
|
| 48 |
+
with yt_dlp.YoutubeDL(ydl_opts) as ydl:
|
| 49 |
+
info = ydl.extract_info(video_url, download=False)
|
| 50 |
+
return info['title']
|
| 51 |
+
except Exception as e:
|
| 52 |
+
logger.error(f"Error fetching title for {video_url}: {str(e)}")
|
| 53 |
+
return f"Unknown Title (Error: {str(e)})"
|
| 54 |
+
|
| 55 |
+
# Function to fetch transcript using youtube_transcript_api
|
| 56 |
+
def get_transcript(video_url: str) -> tuple[Optional[str], Optional[str]]:
|
| 57 |
+
try:
|
| 58 |
+
video_id = video_url.split("v=")[-1].split("&")[0]
|
| 59 |
+
logger.info(f"Fetching transcript for video_id: {video_id}")
|
| 60 |
+
transcript = YouTubeTranscriptApi.get_transcript(video_id)
|
| 61 |
+
text = " ".join([entry["text"] for entry in transcript])
|
| 62 |
+
logger.info(f"Transcript fetched for {video_url}: {text[:100]}...")
|
| 63 |
+
return text, None
|
| 64 |
+
except Exception as e:
|
| 65 |
+
logger.error(f"Error fetching transcript for {video_url}: {str(e)}")
|
| 66 |
+
return None, f"Error fetching transcript for {video_url}: {str(e)}"
|
| 67 |
+
|
| 68 |
+
# Function to summarize text using ChatGPT 4o-mini in bullet points
|
| 69 |
+
def summarize_text(text: str, prompt_prefix: str = "Summarize the following transcript in 2-4 bullet points:") -> str:
|
| 70 |
+
try:
|
| 71 |
+
response = client.chat.completions.create(
|
| 72 |
+
model="gpt-4o-mini",
|
| 73 |
+
messages=[
|
| 74 |
+
{"role": "system", "content": "You are a helpful assistant that summarizes text concisely in bullet point format."},
|
| 75 |
+
{"role": "user", "content": f"{prompt_prefix}\n\n{text}"}
|
| 76 |
+
],
|
| 77 |
+
max_tokens=300,
|
| 78 |
+
temperature=0.7
|
| 79 |
+
)
|
| 80 |
+
summary = response.choices[0].message.content.strip()
|
| 81 |
+
logger.info(f"Generated summary: {summary}")
|
| 82 |
+
return summary
|
| 83 |
+
except Exception as e:
|
| 84 |
+
logger.error(f"Error summarizing text: {str(e)}")
|
| 85 |
+
return f"Error summarizing text: {str(e)}"
|
| 86 |
+
|
| 87 |
+
# LangGraph Nodes
|
| 88 |
+
def fetch_transcripts(state: SummarizerState) -> SummarizerState:
|
| 89 |
+
state["transcripts"] = []
|
| 90 |
+
state["errors"] = []
|
| 91 |
+
for url in state["urls"]:
|
| 92 |
+
transcript, error = get_transcript(url)
|
| 93 |
+
if error:
|
| 94 |
+
state["errors"].append(error)
|
| 95 |
+
state["transcripts"].append("")
|
| 96 |
+
else:
|
| 97 |
+
state["transcripts"].append(transcript)
|
| 98 |
+
logger.info(f"Fetched transcripts: {state['transcripts']}")
|
| 99 |
+
return state
|
| 100 |
+
|
| 101 |
+
def generate_individual_summaries(state: SummarizerState) -> SummarizerState:
|
| 102 |
+
state["summaries"] = []
|
| 103 |
+
for url, title, transcript in zip(state["urls"], state["titles"], state["transcripts"]):
|
| 104 |
+
if not transcript:
|
| 105 |
+
state["summaries"].append(f"**Summary for '{title}' ({url})**:\nNo transcript available.")
|
| 106 |
+
else:
|
| 107 |
+
summary = summarize_text(transcript, "Summarize the following transcript in 2-4 bullet points:")
|
| 108 |
+
state["summaries"].append(f"**Summary for '{title}' ({url})**:\n{summary}")
|
| 109 |
+
logger.info(f"Individual summaries: {state['summaries']}")
|
| 110 |
+
return state
|
| 111 |
+
|
| 112 |
+
def generate_master_summary(state: SummarizerState) -> SummarizerState:
|
| 113 |
+
valid_transcripts = [t for t in state["transcripts"] if t]
|
| 114 |
+
if valid_transcripts:
|
| 115 |
+
combined_text = " ".join(valid_transcripts)
|
| 116 |
+
master_prompt = "Provide a concise summary in 2-4 bullet points that combines key insights from the following transcripts, highlighting common themes or differences:"
|
| 117 |
+
state["master_summary"] = summarize_text(combined_text, master_prompt)
|
| 118 |
+
else:
|
| 119 |
+
state["master_summary"] = "No valid transcripts available for master summary."
|
| 120 |
+
logger.info(f"Master summary: {state['master_summary']}")
|
| 121 |
+
return state
|
| 122 |
+
|
| 123 |
+
def format_output(state: SummarizerState) -> SummarizerState:
|
| 124 |
+
state["master_output"] = f"**Master Summary**:\n{state['master_summary']}"
|
| 125 |
+
state["individual_output"] = "\n\n".join(state["summaries"])
|
| 126 |
+
state["error_output"] = "\n\n".join(state["errors"]) if state["errors"] else "All summaries generated successfully."
|
| 127 |
+
logger.info(f"Formatted output - Master: {state['master_output']}, Individual: {state['individual_output']}, Errors: {state['error_output']}")
|
| 128 |
+
return state
|
| 129 |
+
|
| 130 |
+
# Build the LangGraph workflow
|
| 131 |
+
workflow = StateGraph(SummarizerState)
|
| 132 |
+
|
| 133 |
+
# Add nodes
|
| 134 |
+
workflow.add_node("fetch_transcripts", fetch_transcripts)
|
| 135 |
+
workflow.add_node("generate_individual_summaries", generate_individual_summaries)
|
| 136 |
+
workflow.add_node("generate_master_summary", generate_master_summary)
|
| 137 |
+
workflow.add_node("format_output", format_output)
|
| 138 |
+
|
| 139 |
+
# Define edges
|
| 140 |
+
workflow.add_edge("fetch_transcripts", "generate_individual_summaries")
|
| 141 |
+
workflow.add_edge("generate_individual_summaries", "generate_master_summary")
|
| 142 |
+
workflow.add_edge("generate_master_summary", "format_output")
|
| 143 |
+
workflow.add_edge("format_output", END)
|
| 144 |
+
|
| 145 |
+
# Set the entry point
|
| 146 |
+
workflow.set_entry_point("fetch_transcripts")
|
| 147 |
+
|
| 148 |
+
# Compile the graph with memory
|
| 149 |
+
memory = MemorySaver()
|
| 150 |
+
app = workflow.compile(checkpointer=memory)
|
| 151 |
+
|
| 152 |
+
# Function to add a URL to the list
|
| 153 |
+
def add_url(url: str, state: Dict) -> tuple[Dict, str, str, str]:
|
| 154 |
+
if not url:
|
| 155 |
+
return state, "Please enter a valid YouTube URL.", "", ""
|
| 156 |
+
if len(state["urls"]) >= 3:
|
| 157 |
+
return state, "Maximum limit of 3 URLs reached.", "", ""
|
| 158 |
+
if url in state["urls"]:
|
| 159 |
+
return state, "This URL has already been added.", "", ""
|
| 160 |
+
title = get_video_title(url)
|
| 161 |
+
state["urls"].append(url)
|
| 162 |
+
state["titles"].append(title)
|
| 163 |
+
url_list = "\n".join([f"{title} ({url})" for title, url in zip(state["titles"], state["urls"])])
|
| 164 |
+
return state, f"Added URL: {url}", url_list, ""
|
| 165 |
+
|
| 166 |
+
# Function to summarize all videos using LangGraph
|
| 167 |
+
def summarize_videos(state: Dict) -> tuple[Dict, str, str, str]:
|
| 168 |
+
if not state["urls"]:
|
| 169 |
+
return state, "No URLs added yet.", "", ""
|
| 170 |
+
|
| 171 |
+
# Initialize state for LangGraph
|
| 172 |
+
initial_state = SummarizerState(
|
| 173 |
+
urls=state["urls"],
|
| 174 |
+
titles=state["titles"],
|
| 175 |
+
transcripts=[],
|
| 176 |
+
summaries=[],
|
| 177 |
+
master_summary="",
|
| 178 |
+
errors=[],
|
| 179 |
+
master_output="",
|
| 180 |
+
individual_output="",
|
| 181 |
+
error_output=""
|
| 182 |
+
)
|
| 183 |
+
|
| 184 |
+
# Run the LangGraph workflow
|
| 185 |
+
thread_id = "summarizer_thread"
|
| 186 |
+
final_state = app.invoke(initial_state, config={"configurable": {"thread_id": thread_id}})
|
| 187 |
+
|
| 188 |
+
# Update the Gradio state with the results
|
| 189 |
+
state["summaries"] = final_state["summaries"]
|
| 190 |
+
state["master_summary"] = final_state["master_summary"]
|
| 191 |
+
|
| 192 |
+
return (
|
| 193 |
+
state,
|
| 194 |
+
final_state["error_output"],
|
| 195 |
+
final_state["master_output"],
|
| 196 |
+
final_state["individual_output"]
|
| 197 |
+
)
|
| 198 |
+
|
| 199 |
+
# Function to clear the state
|
| 200 |
+
def clear_state() -> tuple[Dict, str, str, str]:
|
| 201 |
+
return {"urls": [], "titles": [], "summaries": [], "master_summary": ""}, "Cleared all URLs and summaries.", "", ""
|
| 202 |
+
|
| 203 |
+
# Gradio interface
|
| 204 |
+
with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
| 205 |
+
gr.Markdown(INSTRUCTIONS)
|
| 206 |
+
|
| 207 |
+
# Split the interface into two columns
|
| 208 |
+
with gr.Row():
|
| 209 |
+
# Left side: URL input and management
|
| 210 |
+
with gr.Column(scale=1):
|
| 211 |
+
gr.Markdown("### Add YouTube URLs")
|
| 212 |
+
url_input = gr.Textbox(label="Enter YouTube URL", placeholder="https://www.youtube.com/watch?v=...")
|
| 213 |
+
add_button = gr.Button("Add URL")
|
| 214 |
+
url_list = gr.Textbox(label="Added URLs", interactive=False, lines=5)
|
| 215 |
+
with gr.Row():
|
| 216 |
+
summarize_button = gr.Button("Summarize")
|
| 217 |
+
clear_button = gr.Button("Clear")
|
| 218 |
+
status_output = gr.Textbox(label="Status")
|
| 219 |
+
|
| 220 |
+
# Right side: Summaries
|
| 221 |
+
with gr.Column(scale=2):
|
| 222 |
+
gr.Markdown("### Summaries")
|
| 223 |
+
master_summary = gr.Markdown(label="Master Summary", min_height=200)
|
| 224 |
+
individual_summaries = gr.Markdown(label="Individual Summaries", min_height=300)
|
| 225 |
+
|
| 226 |
+
# State to track URLs, titles, and summaries
|
| 227 |
+
state = gr.State(value={"urls": [], "titles": [], "summaries": [], "master_summary": ""})
|
| 228 |
+
|
| 229 |
+
# Event handlers
|
| 230 |
+
add_button.click(
|
| 231 |
+
fn=add_url,
|
| 232 |
+
inputs=[url_input, state],
|
| 233 |
+
outputs=[state, status_output, url_list, url_input]
|
| 234 |
+
)
|
| 235 |
+
summarize_button.click(
|
| 236 |
+
fn=summarize_videos,
|
| 237 |
+
inputs=[state],
|
| 238 |
+
outputs=[state, status_output, master_summary, individual_summaries]
|
| 239 |
+
)
|
| 240 |
+
clear_button.click(
|
| 241 |
+
fn=clear_state,
|
| 242 |
+
inputs=[],
|
| 243 |
+
outputs=[state, status_output, master_summary, individual_summaries]
|
| 244 |
+
)
|
| 245 |
+
|
| 246 |
+
# Launch the app
|
| 247 |
+
demo.launch(share=False)
|