Spaces:
Build error
Build error
| import gradio as gr | |
| import openai, os | |
| from langchain.chains import LLMChain, RetrievalQA | |
| from langchain.chat_models import ChatOpenAI | |
| from langchain.document_loaders import PyPDFLoader | |
| 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 = """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. """ | |
| llm_template = "Answer the question at the end. " + template + "Question: {question} Helpful Answer: " | |
| rag_template = "Use the following pieces of context to answer the question at the end. " + template + "{context} Question: {question} Helpful Answer: " | |
| LLM_CHAIN_PROMPT = PromptTemplate(input_variables = ["question"], | |
| template = llm_template) | |
| RAG_CHAIN_PROMPT = PromptTemplate(input_variables = ["context", "question"], | |
| template = rag_template) | |
| CHROMA_DIR = "/data/chroma" | |
| YOUTUBE_DIR = "/data/youtube" | |
| YOUTUBE_URL_1 = "https://www.youtube.com/watch?v=--khbXchTeE" | |
| YOUTUBE_URL_2 = "https://www.youtube.com/watch?v=hdhZwyf24mE" | |
| YOUTUBE_URL_3 = "https://www.youtube.com/watch?v=vw-KWfKwvTQ" | |
| YOUTUBE_URL_4 = "https://www.youtube.com/watch?v=kiHpqXNCPj8" | |
| YOUTUBE_URL_5 = "https://www.youtube.com/shorts/3x95mw35dJY" | |
| YOUTUBE_URL_6 = "https://www.youtube.com/shorts/zg-DS23wq0c" | |
| YOUTUBE_URL_7 = "https://www.youtube.com/shorts/cS4fyhKZ8bQ" | |
| PDF_URL = "https://arxiv.org/pdf/2303.08774.pdf" | |
| MODEL_NAME = "gpt-4" | |
| def invoke(openai_api_key, use_rag, prompt): | |
| llm = ChatOpenAI(model_name = MODEL_NAME, | |
| openai_api_key = openai_api_key, | |
| temperature = 0) | |
| if (use_rag): | |
| # Document loading, splitting, and storage | |
| docs = [] | |
| #loader = GenericLoader(YoutubeAudioLoader([YOUTUBE_URL_1, | |
| # YOUTUBE_URL_2, | |
| # YOUTUBE_URL_3, | |
| # YOUTUBE_URL_4, | |
| # YOUTUBE_URL_5, | |
| # YOUTUBE_URL_6, | |
| # YOUTUBE_URL_7], YOUTUBE_DIR), | |
| # OpenAIWhisperParser()) | |
| #docs = loader.load() | |
| ###docs.extend(loader.load()) | |
| loader = PyPDFLoader(PDF_URL) | |
| docs.extend(loader.load()) | |
| # | |
| text_splitter = RecursiveCharacterTextSplitter(chunk_overlap = 150, | |
| chunk_size = 1500) | |
| splits = text_splitter.split_documents(docs) | |
| vector_db = Chroma.from_documents(documents = splits, | |
| embedding = OpenAIEmbeddings(), | |
| persist_directory = CHROMA_DIR) | |
| # Document retrieval | |
| vector_db = Chroma(embedding_function = OpenAIEmbeddings(), | |
| persist_directory = CHROMA_DIR) | |
| rag_chain = RetrievalQA.from_chain_type(llm, | |
| chain_type_kwargs = {"prompt": RAG_CHAIN_PROMPT}, | |
| retriever = vector_db.as_retriever(search_kwargs = {"k": 3}), | |
| return_source_documents = True) | |
| result = rag_chain({"query": prompt}) | |
| result = result["result"] | |
| else: | |
| chain = LLMChain(llm = llm, prompt = LLM_CHAIN_PROMPT) | |
| result = chain.run({"question": prompt}) | |
| return result | |
| description = """<strong>Overview:</strong> The app demonstrates how to use a Large Language Model (LLM) with Retrieval Augmented Generation (RAG) on external data | |
| (YouTube videos, PDFs, URLs, or other <a href='https://raw.githubusercontent.com/bstraehle/ai-ml-dl/c38b224c196fc984aab6b6cc6bdc666f8f4fbcff/langchain/document-loaders.png'>data sources</a>).\n\n | |
| <strong>Instructions:</strong> Enter an OpenAI API key and perform LLM use cases (semantic search, summarization, translation, etc.) on | |
| <strong>YouTube videos about GPT-4</strong>, created after its training cutoff. | |
| <ul style="list-style-type:square;"> | |
| <li>Set "Retrieval Augmented Generation" to "<strong>False</strong>" and submit prompt "What is GPT-4?" The LLM <strong>without</strong> RAG does not know the answer.</li> | |
| <li>Set "Retrieval Augmented Generation" to "<strong>True</strong>" and submit prompt "What is GPT-4?" The LLM <strong>with</strong> RAG knows the answer.</li> | |
| <li>Experiment with prompts, e.g. "What is GPT-4 in one sentence in German", "List pros and cons of GPT-4", or "Write a Python program to call the GPT-4 API".</li> | |
| </ul>\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.Radio([True, False], label="Retrieval Augmented Generation", value = False), | |
| 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() |