File size: 4,389 Bytes
7d6d701 9621cc7 7d6d701 6f02f68 1ad0dcf 6f02f68 1ad0dcf 7d6d701 9ed9edc 7d6d701 c8f85cc dffbc3f 52f3a4a 6f02f68 752918c b610816 fcdc77d 6553dbd b12409c 9960268 9ed9edc 9960268 7cacaa1 5917f38 57e6710 8cffc38 9960268 8cffc38 4f5dd89 aa58995 4f5dd89 03e3607 6f02f68 7d6d701 42b515d c11aec6 42b515d 954479c 03fde88 96950c8 42b515d 423b214 290e7c0 7d6d701 2cf5d84 9ed9edc 908ded3 7d6d701 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 |
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())
#openai.api_key = os.environ["OPENAI_API_KEY"]
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(search_kwargs = {"k": 5})), return_source_documents = True, chain_type_kwargs = {"prompt": QA_CHAIN_PROMPT})
result = qa_chain({"query": prompt})
print(result)
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 run and then to "False". 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() |