import os from typing import Any import gradio as gr import openai import pandas as pd from IPython.display import Markdown, display from langchain.document_loaders import PyPDFLoader from langchain.embeddings import OpenAIEmbeddings from langchain.indexes import VectorstoreIndexCreator from langchain.text_splitter import CharacterTextSplitter from langchain.text_splitter import RecursiveCharacterTextSplitter from langchain.llms import OpenAI from langchain.vectorstores import DocArrayInMemorySearch from uuid import uuid4 css_style = """ .gradio-container { font-family: "IBM Plex Mono"; } """ class myClass: def __init__(self) -> None: self.openapi = "" self.valid_key = False self.docs_ready = False self.status = "⚠️Waiting for documents and key⚠️" self.uuid = uuid4() pass def check_status(self): if self.docs_ready and self.valid_key: out = "✨Ready✨" elif self.docs_ready: out = "⚠️Waiting for key⚠️" elif self.valid_key: out = "⚠️Waiting for documents⚠️" else: out = "⚠️Waiting for documents and key⚠️" self.status = out def validate_key(self, myin): assert isinstance(myin, str) self.valid_key = True self.openai_api_key = myin.strip() self.embedding = OpenAIEmbeddings(openai_api_key=self.openai_api_key) self.llm = OpenAI(openai_api_key=self.openai_api_key) self.check_status() return [self.status] def request_pathname(self, files, data): if files is None: self.docs_ready = False self.check_status() return ( pd.DataFrame(data, columns=["filepath", "citation string", "key"]), self.status, ) for file in files: # make sure we're not duplicating things in the dataset if file.name in [x[0] for x in data]: continue data.append([file.name, None, None]) mydataset = pd.DataFrame(data, columns=["filepath", "citation string", "key"]) validation_button = self.validate_dataset(mydataset) return mydataset, validation_button def validate_dataset(self, dataset): self.docs_ready = dataset.iloc[-1, 0] != "" self.dataset = dataset self.check_status() if self.status == "✨Ready✨": self.get_index() return self.status def get_index(self): if self.docs_ready and self.valid_key: # os.environ["OPENAI_API_KEY"] = self.openai_api_key # myfile = "Angela Merkel - Wikipedia.pdf" # loader = PyPDFLoader(file_path=myfile) loaders = [PyPDFLoader(f) for f in self.dataset["filepath"]] self.index = VectorstoreIndexCreator( vectorstore_cls=DocArrayInMemorySearch, embedding=self.embedding, text_splitter = RecursiveCharacterTextSplitter( # Set a really small chunk size, just to show. chunk_size = 1000, chunk_overlap = 20, length_function = len, separators="." ) ).from_loaders(loaders=loaders) # del os.environ["OPENAI_API_KEY"] pass def do_ask(self, question): # os.environ["OPENAI_API_KEY"] = self.openai_api_key # openai.api_key = self.openai_api_key if self.status == "✨Ready✨": # os.environ["OPENAI_API_KEY"] = self.openai_api_key response = self.index.query(question=question, llm=self.llm) # del os.environ["OPENAI_API_KEY"] yield response pass def validate_key(myInstance: myClass, openai_api_key): if myInstance is None: myInstance = myClass() out = myInstance.validate_key(openai_api_key) return myInstance, *out def request_pathname(myInstance: myClass, files, data): if myInstance is None: myInstance = myClass() out = myInstance.request_pathname(files, data) return myInstance, *out def do_ask(myInstance: myClass, question): out = myInstance.do_ask(question) return myInstance, *out with gr.Blocks(css=css_style) as demo: myInstance = gr.State() openai_api_key = gr.State("") docs = gr.State() data = gr.State([]) index = gr.State() gr.Markdown( """ # Document Question and Answer *By D8a.ai* Idea based on https://huggingface.co/spaces/whitead/paper-qa Significant advances in langchain have made it possible to simplify the code. This tool allows you to ask questions of your uploaded text, PDF documents. It uses OpenAI's GPT models, so you need to enter your API key below. This tool is under active development and currently uses a lot of tokens - up to 10,000 for a single query. This is $0.10-0.20 per query, so please be careful! * [langchain](https://github.com/hwchase17/langchain) is the main library this tool utilizes. 1. Enter API Key ([What is that?](https://platform.openai.com/account/api-keys)) 2. Upload your documents 3. Ask questions """ ) openai_api_key = gr.Textbox( label="OpenAI API Key", placeholder="sk-...", type="password" ) with gr.Tab("File upload"): uploaded_files = gr.File( label="Upload your pdf Dokument", file_count="multiple" ) with gr.Accordion("See Docs:", open=False): dataset = gr.Dataframe( headers=["filepath", "citation string", "key"], datatype=["str", "str", "str"], col_count=(3, "fixed"), interactive=False, label="Documents and Citations", overflow_row_behaviour="paginate", max_rows=5, ) buildb = gr.Textbox( "⚠️Waiting for documents and key...", label="Status", interactive=False, show_label=True, max_lines=1, ) query = gr.Textbox(placeholder="Enter your question here...", label="Question") ask = gr.Button("Ask Question") answer = gr.Markdown(label="Answer") openai_api_key.change( validate_key, inputs=[myInstance, openai_api_key], outputs=[myInstance, buildb] ) uploaded_files.change( request_pathname, inputs=[myInstance, uploaded_files, data], outputs=[myInstance, dataset, buildb], ) ask.click( do_ask, inputs=[myInstance, query], outputs=[myInstance, answer], ) demo.queue(concurrency_count=20) demo.launch(show_error=True)