|
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) |
|
|
|
CHROMA_DIR = "docs/chroma/" |
|
YOUTUBE_DIR = "docs/youtube/" |
|
|
|
MODEL_NAME = "gpt-4" |
|
|
|
def invoke(openai_api_key, youtube_url, process_video, prompt): |
|
openai.api_key = openai_api_key |
|
if (process_video): |
|
if (os.path.isdir(CHROMA_DIR)): |
|
shutil.rmtree(CHROMA_DIR) |
|
if (os.path.isdir(YOUTUBE_DIR)): |
|
shutil.rmtree(YOUTUBE_DIR) |
|
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) |
|
vector_db = Chroma.from_documents(documents = splits, embedding = OpenAIEmbeddings(), persist_directory = CHROMA_DIR) |
|
else: |
|
vector_db = Chroma(persist_directory = CHROMA_DIR, embedding_function = OpenAIEmbeddings()) |
|
llm = ChatOpenAI(model_name = MODEL_NAME, temperature = 0) |
|
qa_chain = RetrievalQA.from_chain_type(llm, retriever = vector_db.as_retriever(), return_source_documents = True, chain_type_kwargs = {"prompt": QA_CHAIN_PROMPT}) |
|
result = qa_chain({"query": prompt}) |
|
return result["result"] |
|
|
|
description = """<strong>Overview:</strong> The app demonstrates how to use a <strong>Large Language Model</strong> (LLM) with <strong>Retrieval Augmented Generation</strong> |
|
(RAG) on external data (YouTube videos in this case, but could be PDFs, URLs, databases, etc.)\n\n |
|
<strong>Instructions:</strong> Enter an OpenAI API key, YouTube URL, and prompt to perform semantic search, sentiment analysis, summarization, |
|
translation, etc. "Process Video" specifies whether or not to perform speech-to-text processing. To ask multiple questions related to the same video, |
|
typically set it to "True" the first time and then to "False". Note that persistence is not guaranteed in the Hugging Face free tier |
|
(the plan is to migrate to AWS S3). The example is a 3:12 min. video about GPT-4 and takes about 20 sec. to process. Try different prompts, for example |
|
"what is gpt-4, answer in german" or "write a poem about gpt-4".\n\n |
|
<strong>Technology:</strong> <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) 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.Radio([True, False], label="Process Video", value = True), gr.Textbox(label = "Prompt", value = "what is gpt-4", lines = 1)], |
|
outputs = [gr.Textbox(label = "Completion", lines = 1)], |
|
title = "Generative AI - LLM & RAG", |
|
description = description) |
|
demo.launch() |
