app.py

import gradio as gr
import pytube
from youtube_transcript_api import YouTubeTranscriptApi as yt
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import os
from langchain import PromptTemplate
from langchain import LLMChain
from langchain_together import Together
import re

# Set the API key with double quotes
os.environ['TOGETHER_API_KEY'] = "d88cb7414e4039a84d2ed63f1b47daaaa4230c4c53a422045d8a30a9a3bc87d8"

def Summary_BART(text):
    checkpoint = "sshleifer/distilbart-cnn-12-6"
    tokenizer = AutoTokenizer.from_pretrained(checkpoint)
    model = AutoModelForSeq2SeqLM.from_pretrained(checkpoint)
    inputs = tokenizer(text, max_length=1024, truncation=True, return_tensors="pt")
    summary_ids = model.generate(inputs["input_ids"])
    summary = tokenizer.batch_decode(summary_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)
    return summary[0]

def YtToQuizz(link, difficulty_level):
    video_id = pytube.extract.video_id(link)
    transcript = yt.get_transcript(video_id)
    data = ""
    for text in transcript:
        data += text.get('text') + " "
    summary = Summary_BART(data)
    
    mcq_template = """
    Generate 10 different multiple-choice questions (MCQs) related to the following summary: {summary}
    The difficulty level of the questions should be: {difficulty_level}
    Please provide the following for each question:
    1. Question
    2. Correct answer
    3. Three plausible incorrect answer options
    4. Format: "Question: <question text>\\nCorrect answer: <correct answer>\\nIncorrect answers: <option1>, <option2>, <option3>"
    """ 
    prompt = PromptTemplate(
        input_variables=['summary', 'difficulty_level'],
        template=mcq_template
    )
    llama3 = Together(model="meta-llama/Llama-3-70b-chat-hf", max_tokens=2500)
    Generated_mcqs = LLMChain(llm=llama3, prompt=prompt)

    response = Generated_mcqs.invoke({
        "summary": summary,
        "difficulty_level": difficulty_level
    })

    response_text = response['text']

    # Extract MCQs
    mcq_pattern = r'Question: (.*?)\nCorrect answer: (.*?)\nIncorrect answers: (.*?)(?:\n|$)'
    mcqs = re.findall(mcq_pattern, response_text, re.DOTALL)

    if len(mcqs) < 10:
        return "Failed to generate 10 complete MCQs. Please try again.", "", ""

    questions_str = ""
    correct_answers_str = ""
    options_str = ""

    for idx, mcq in enumerate(mcqs[:10]):
        question, correct_answer, incorrect_answers = mcq
        incorrect_answers = incorrect_answers.split(', ')
        questions_str += f"Q{idx+1}: {question}, "
        correct_answers_str += f"Q{idx+1}: {correct_answer}, "
        options_str += f"Q{idx+1}: A) {correct_answer}, B) {incorrect_answers[0]}, C) {incorrect_answers[1]}, D) {incorrect_answers[2]}, "

    # Removing the trailing comma and space
    questions_str = questions_str.rstrip(", ")
    correct_answers_str = correct_answers_str.rstrip(", ")
    options_str = options_str.rstrip(", ")

    return questions_str, correct_answers_str, options_str

def main(link, difficulty_level):
    return YtToQuizz(link, difficulty_level)

iface = gr.Interface(
    fn=main,
    inputs=[
        gr.components.Textbox(lines=2, placeholder="Enter YouTube video link"),
        gr.components.Dropdown(["Easy", "Medium", "Hard"], label="Select difficulty level:")
    ],
    outputs=[
        gr.components.Textbox(label="MCQs Statements", lines=20),
        gr.components.Textbox(label="Correct Answers", lines=10),
        gr.components.Textbox(label="Options", lines=30)
    ],
    title="YouTube Video Subtitle to MCQs Quiz",
    description="Generate MCQs from YouTube video subtitles"
)

if __name__ == '__main__':
    iface.launch()