|
import gradio as gr |
|
import shutil, openai, os |
|
|
|
from langchain.chains import RetrievalQA |
|
from langchain.chat_models import ChatOpenAI |
|
from langchain.document_loaders.blob_loaders.youtube_audio import YoutubeAudioLoader |
|
from langchain.document_loaders.generic import GenericLoader |
|
from langchain.document_loaders.parsers import OpenAIWhisperParser |
|
from langchain.embeddings.openai import OpenAIEmbeddings |
|
from langchain.prompts import PromptTemplate |
|
from langchain.text_splitter import RecursiveCharacterTextSplitter |
|
from langchain.vectorstores import Chroma |
|
|
|
from dotenv import load_dotenv, find_dotenv |
|
_ = load_dotenv(find_dotenv()) |
|
|
|
|
|
|
|
template = """Use the following pieces of context to answer the question at the end. If you don't know the answer, just say that you don't know, don't try to make up |
|
an answer. Keep the answer as concise as possible. Always say "🔥 Thanks for using the app, Bernd Straehle." at the end of the answer. |
|
{context} Question: {question} Helpful Answer: """ |
|
|
|
QA_CHAIN_PROMPT = PromptTemplate(input_variables = ["context", "question"], template = template) |
|
|
|
print(0) |
|
|
|
qa_chain = None |
|
|
|
def invoke(openai_api_key, youtube_url, prompt): |
|
openai.api_key = openai_api_key |
|
if (os.path.isdir("docs/chroma/") == False): |
|
print(1) |
|
youtube_dir = "docs/youtube/" |
|
loader = GenericLoader(YoutubeAudioLoader([youtube_url], youtube_dir), OpenAIWhisperParser()) |
|
docs = loader.load() |
|
text_splitter = RecursiveCharacterTextSplitter(chunk_size = 1500, chunk_overlap = 150) |
|
splits = text_splitter.split_documents(docs) |
|
chroma_dir = "docs/chroma/" |
|
vectordb = Chroma.from_documents(documents = splits, embedding = OpenAIEmbeddings(), persist_directory = chroma_dir) |
|
llm = ChatOpenAI(model_name = "gpt-4", temperature = 0) |
|
global qa_chain = RetrievalQA.from_chain_type(llm, retriever = vectordb.as_retriever(), return_source_documents = True, chain_type_kwargs = {"prompt": QA_CHAIN_PROMPT}) |
|
print(2) |
|
result = global qa_chain({"query": prompt}) |
|
shutil.rmtree(youtube_dir) |
|
|
|
return result["result"] |
|
|
|
description = """The app demonstrates how to use a <strong>Large Language Model</strong> (LLM) with <strong>Retrieval Augmented Generation</strong> (RAG) on external data. |
|
Enter an OpenAI API key, YouTube URL (external data), and prompt to perform semantic search, sentiment analysis, summarization, translation, etc.\n\n |
|
Implementation: <a href='https://www.gradio.app/'>Gradio</a> UI using <a href='https://platform.openai.com/'>OpenAI</a> API |
|
via AI-first <a href='https://www.langchain.com/'>LangChain</a> toolkit with <a href='https://openai.com/research/whisper'>Whisper</a> (speech to text) |
|
and <a href='https://openai.com/research/gpt-4'>GPT-4</a> (LLM use cases) foundation models as well as AI-native |
|
<a href='https://www.trychroma.com/'>Chroma</a> embedding database.""" |
|
|
|
gr.close_all() |
|
demo = gr.Interface(fn=invoke, |
|
inputs = [gr.Textbox(label = "OpenAI API Key", value = "sk-", lines = 1), gr.Textbox(label = "YouTube URL", value = "https://www.youtube.com/watch?v=--khbXchTeE", lines = 1), gr.Textbox(label = "Prompt", value = "GPT-4 human level performance", lines = 1)], |
|
outputs = [gr.Textbox(label = "Completion", lines = 1)], |
|
title = "Generative AI - LLM & RAG", |
|
description = description) |
|
demo.launch() |
