app.py

import gradio as gr
from langchain_community.document_loaders import YoutubeLoader
from langchain.text_splitter import TokenTextSplitter
from groq import Groq
import os

client = Groq(
    api_key=os.environ.get("api_key"),
)
max_textboxes = 5

def process_youtube_url(url, language):
    try:
        loader = YoutubeLoader.from_youtube_url(
            youtube_url=url,
            add_video_info=True,
            language=[language],
        )
        docs = loader.load()
        text = str(docs)
        
        # embeddings = OpenAIEmbeddings()
        token_count = len(text)
        
        text_splitter = TokenTextSplitter(chunk_size=30_000, chunk_overlap=0)
        chunks = text_splitter.split_text(text)
        
        output_textboxes = [chunk for i, chunk in enumerate(chunks)]
        output_textboxes += ["" for _ in range(max_textboxes - len(chunks))]

        completion = client.chat.completions.create(
            model="llama3-70b-8192",
            messages=[
                {
                    "role": "system",
                    "content": "lang:日本語 あなたは日本語話者の解説系Youtuberです。"
                },
                {
                    "role": "user",
                    "content": f"lang:日本語　日本語で次のtranscriptを解説して\n\n## trascript \n{text}"
                },
                {
                    "role": "assisnat",
                    "content": f"この動画は、"
                },
            ],
            temperature=0.7,
            max_tokens=1024,
            top_p=1,
            stream=True,
            stop=None,
        )
        summirizedtext = ""
        for chunk in completion:
            summirizedtext += chunk.choices[0].delta.content or ""
            # print(chunk.choices[0].delta.content or "", end="")
        
            yield *output_textboxes, [], text, token_count,summirizedtext
    
    except Exception as e:
        error_msg = str(e)
        available_languages = extract_available_languages(error_msg)
        recommended_language = extract_recommended_language(error_msg)
        yield *[error_msg for _ in range(max_textboxes)], available_languages, recommended_language, 0,""


def extract_available_languages(error_msg):
    languages = []
    generated_section = False
    for line in error_msg.split("\n"):
        if line.startswith("(GENERATED)"):
            generated_section = True
        elif generated_section and line.startswith(" - "):
            lang_code, lang_name = line[3:].split(" (", 1)
            languages.append(f"{lang_name[:-1]} ({lang_code})")
    return languages

def extract_recommended_language(error_msg):
    generated_section = False
    for line in error_msg.split("\n"):
        if line.startswith("(GENERATED)"):
            generated_section = True
        elif generated_section and line.startswith(" - ") and "[TRANSLATABLE]" in line:
            lang_code, lang_name = line[3:].split(" (", 1)
            return f"{lang_name[:-1]} ({lang_code})"
    return ""

iface = gr.Interface(
    fn=process_youtube_url,
    inputs=[
        gr.Textbox(label="YouTube URL", placeholder="https://youtu.be/example"),
        gr.Dropdown(label="Language",value="ja",choices=["en","en-US", "ja", "fr","de","it"],allow_custom_value=True),
    ],
    outputs=
        [gr.Textbox(label=f"chunk{ind}",show_copy_button=True,max_lines=5) for ind in range(max_textboxes)]
    +[
        gr.Dropdown(label="Available Languages", allow_custom_value=True),
        gr.Textbox(label="Recommended Language",show_copy_button=True),
        gr.Number(label="Character Count"),
        gr.Markdown(label='summirized output'),
    ],
    live=True,
    examples = [["https://youtu.be/6Af6b_wyiwI?si=zqD9-kjw24lpRJw3","ja"],["https://youtu.be/9kxL9Cf46VM?si=ADgUmDXb6riA-lgb","ja"]],
    title="YouTube Transcript Loader",
    description="Enter a YouTube URL and select the language to load the transcript using LangChain's YoutubeLoader.[buy me a coffee](https://www.buymeacoffee.com/regulusle04)",
)

if __name__ == "__main__":
    iface.queue()
    iface.launch(share=True)