import json import logging import os import shutil import sys import uuid from json import JSONDecodeError from pathlib import Path import pandas as pd import pinecone import streamlit as st from annotated_text import annotation from haystack import Document from haystack.document_stores import PineconeDocumentStore from haystack.nodes import ( DocxToTextConverter, EmbeddingRetriever, FARMReader, FileTypeClassifier, PDFToTextConverter, PreProcessor, TextConverter, ) from haystack.pipelines import ExtractiveQAPipeline, Pipeline from markdown import markdown from sentence_transformers import SentenceTransformer import openai # get API key from top-right dropdown on OpenAI website openai.api_key = st.secrets["OPENAI_API_KEY"] index_name = "qa_demo" # connect to pinecone environment pinecone.init( api_key=st.secrets["pinecone_apikey"], environment="us-east1-gcp" ) index_name = "qa-demo" embed_model = "text-embedding-ada-002" preprocessor = PreProcessor( clean_empty_lines=True, clean_whitespace=True, clean_header_footer=False, split_by="word", split_length=100, split_respect_sentence_boundary=True ) file_type_classifier = FileTypeClassifier() text_converter = TextConverter() pdf_converter = PDFToTextConverter() docx_converter = DocxToTextConverter() # check if the abstractive-question-answering index exists if index_name not in pinecone.list_indexes(): # create the index if it does not exist pinecone.create_index( index_name, dimension=1536, metric="cosine" ) # connect to abstractive-question-answering index we created index = pinecone.Index(index_name) FILE_UPLOAD_PATH= "./data/uploads/" os.makedirs(FILE_UPLOAD_PATH, exist_ok=True) # @st.cache def create_doc_store(): document_store = PineconeDocumentStore( api_key= st.secrets["pinecone_apikey"], index=index_name, similarity="cosine", embedding_dim=768, metadata_config={ 'indexed': ['filename'] } ) return document_store # @st.cache # def create_pipe(document_store): # retriever = EmbeddingRetriever( # document_store=document_store, # embedding_model="sentence-transformers/multi-qa-mpnet-base-dot-v1", # model_format="sentence_transformers", # ) # reader = FARMReader(model_name_or_path="deepset/roberta-base-squad2", use_gpu=False) # pipe = ExtractiveQAPipeline(reader, retriever) # return pipe limit = 3750 def retrieve(query): res = openai.Embedding.create( input=[query], engine=embed_model ) # retrieve from Pinecone xq = res['data'][0]['embedding'] # get relevant contexts res = index.query(xq, top_k=3, include_metadata=True) contexts = [ x['metadata']['text'] for x in res['matches'] ] # build our prompt with the retrieved contexts included prompt_start = ( "Answer the question based on the context below.\n\n"+ "Context:\n" ) prompt_end = ( f"\n\nQuestion: {query}\nAnswer:" ) # append contexts until hitting limit for i in range(1, len(contexts)): if len("\n\n---\n\n".join(contexts[:i])) >= limit: prompt = ( prompt_start + "\n\n---\n\n".join(contexts[:i-1]) + prompt_end ) break elif i == len(contexts)-1: prompt = ( prompt_start + "\n\n---\n\n".join(contexts) + prompt_end ) return prompt, contexts # first let's make it simpler to get answers def complete(prompt): # query text-davinci-003 res = openai.Completion.create( engine='text-davinci-003', prompt=prompt, temperature=0, max_tokens=400, top_p=1, frequency_penalty=0, presence_penalty=0, stop=None ) return res['choices'][0]['text'].strip() def query(pipe, question, top_k_reader, top_k_retriever): # first we retrieve relevant items from Pinecone query_with_contexts, contexts = retrieve(question) return complete(query_with_contexts), contexts document_store = create_doc_store() # pipe = create_pipe(document_store) retriever_model = "sentence-transformers/multi-qa-mpnet-base-dot-v1" retriever = EmbeddingRetriever( document_store=document_store, embedding_model=retriever_model, model_format="sentence_transformers", ) # load the retriever model from huggingface model hub # sentence_encoder = SentenceTransformer(retriever_model) # reader = FARMReader(model_name_or_path="deepset/roberta-base-squad2", use_gpu=False) # pipe = ExtractiveQAPipeline(reader, retriever) # now query text-davinci-003 WITHOUT context indexing_pipeline_with_classification = Pipeline() indexing_pipeline_with_classification.add_node( component=file_type_classifier, name="FileTypeClassifier", inputs=["File"] ) indexing_pipeline_with_classification.add_node( component=text_converter, name="TextConverter", inputs=["FileTypeClassifier.output_1"] ) indexing_pipeline_with_classification.add_node( component=pdf_converter, name="PdfConverter", inputs=["FileTypeClassifier.output_2"] ) indexing_pipeline_with_classification.add_node( component=docx_converter, name="DocxConverter", inputs=["FileTypeClassifier.output_4"] ) indexing_pipeline_with_classification.add_node( component=preprocessor, name="Preprocessor", inputs=["TextConverter", "PdfConverter", "DocxConverter"], ) def set_state_if_absent(key, value): if key not in st.session_state: st.session_state[key] = value # Adjust to a question that you would like users to see in the search bar when they load the UI: DEFAULT_QUESTION_AT_STARTUP = os.getenv("DEFAULT_QUESTION_AT_STARTUP", "My blog post discusses remote work. Give me statistics.") DEFAULT_ANSWER_AT_STARTUP = os.getenv("DEFAULT_ANSWER_AT_STARTUP", "7% more remote workers have been at their current organization for 5 years or fewer") # Sliders DEFAULT_DOCS_FROM_RETRIEVER = int(os.getenv("DEFAULT_DOCS_FROM_RETRIEVER", "3")) DEFAULT_NUMBER_OF_ANSWERS = int(os.getenv("DEFAULT_NUMBER_OF_ANSWERS", "3")) st.set_page_config(page_title="Haystack Demo", page_icon="https://haystack.deepset.ai/img/HaystackIcon.png") # Persistent state set_state_if_absent("question", DEFAULT_QUESTION_AT_STARTUP) set_state_if_absent("answer", DEFAULT_ANSWER_AT_STARTUP) set_state_if_absent("results", None) # Small callback to reset the interface in case the text of the question changes def reset_results(*args): st.session_state.answer = None st.session_state.results = None st.session_state.raw_json = None # Title st.write("# GPT3 and Langchain Demo") st.markdown( """ This demo takes its data from the documents uploaded to the Pinecone index through this app. \n Ask any question from the uploaded documents and Pinecone will retrieve the context for answers and GPT3 will answer them using the retrieved context. \n *Note: do not use keywords, but full-fledged questions.* The demo is not optimized to deal with keyword queries and might misunderstand you. """, unsafe_allow_html=True, ) # Sidebar st.sidebar.header("Options") st.sidebar.write("## File Upload:") data_files = st.sidebar.file_uploader( "upload", type=["pdf", "txt", "docx"], accept_multiple_files=True, label_visibility="hidden" ) ALL_FILES = [] META_DATA = [] for data_file in data_files: # Upload file if data_file: file_path = Path(FILE_UPLOAD_PATH) / f"{uuid.uuid4().hex}_{data_file.name}" with open(file_path, "wb") as f: f.write(data_file.getbuffer()) ALL_FILES.append(file_path) st.sidebar.write(str(data_file.name) + "    ✅ ") META_DATA.append({"filename":data_file.name}) if len(ALL_FILES) > 0: # document_store.update_embeddings(retriever, update_existing_embeddings=False) docs = indexing_pipeline_with_classification.run(file_paths=ALL_FILES, meta=META_DATA)["documents"] index_name = "qa_demo" # we will use batches of 64 batch_size = 64 # docs = docs['documents'] with st.spinner( "🧠    Performing indexing of uplaoded documents... \n " ): for i in range(0, len(docs), batch_size): # find end of batch i_end = min(i+batch_size, len(docs)) # extract batch batch = [doc.content for doc in docs[i:i_end]] # generate embeddings for batch try: res = openai.Embedding.create(input=texts, engine=embed_model) except: done = False while not done: sleep(5) try: res = openai.Embedding.create(input=texts, engine=embed_model) done = True except: pass embeds = [record['embedding'] for record in res['data']] # get metadata meta = [doc.meta for doc in docs[i:i_end]] # create unique IDs ids = [doc.id for doc in docs[i:i_end]] # add all to upsert list to_upsert = list(zip(ids, emb, meta)) # upsert/insert these records to pinecone _ = index.upsert(vectors=to_upsert) # top_k_reader = st.sidebar.slider( # "Max. number of answers", # min_value=1, # max_value=10, # value=DEFAULT_NUMBER_OF_ANSWERS, # step=1, # on_change=reset_results, # ) # top_k_retriever = st.sidebar.slider( # "Max. number of documents from retriever", # min_value=1, # max_value=10, # value=DEFAULT_DOCS_FROM_RETRIEVER, # step=1, # on_change=reset_results, # ) # data_files = st.file_uploader( # "upload", type=["csv"], accept_multiple_files=True, label_visibility="hidden" # ) # for data_file in data_files: # # Upload file # if data_file: # raw_json = upload_doc(data_file) question = st.text_input( value=st.session_state.question, max_chars=100, on_change=reset_results, label="question", label_visibility="hidden", ) col1, col2 = st.columns(2) col1.markdown("", unsafe_allow_html=True) col2.markdown("", unsafe_allow_html=True) # Run button run_pressed = col1.button("Run") if run_pressed: run_query = ( run_pressed or question != st.session_state.question ) # Get results for query if run_query and question: reset_results() st.session_state.question = question with st.spinner( "🧠    Performing neural search on documents... \n " ): try: st.session_state.results = query( pipe, question, top_k_reader=None, top_k_retriever=None ) except JSONDecodeError as je: st.error("👓    An error occurred reading the results. Is the document store working?") except Exception as e: logging.exception(e) if "The server is busy processing requests" in str(e) or "503" in str(e): st.error("🧑‍🌾    All our workers are busy! Try again later.") else: st.error(f"🐞    An error occurred during the request. {str(e)}") if st.session_state.results: st.write("## Results:") for result,contexts in st.session_state.results: # answer, context = result.answer, result.context # start_idx = context.find(answer) # end_idx = start_idx + len(answer) # Hack due to this bug: https://github.com/streamlit/streamlit/issues/3190 try: # source = f"[{result.meta['Title']}]({result.meta['link']})" # st.write( # markdown(f'**Source:** {source} \n {context[:start_idx] } {str(annotation(answer, "ANSWER", "#8ef"))} {context[end_idx:]} \n '), # unsafe_allow_html=True, # ) st.write( markdown(f"Answer: {result} \n Extracted from context {contexts}"), unsafe_allow_html=True, ) except: # filename = result.meta.get('filename', "") # st.write( # markdown(f'From file: {filename} \n {context[:start_idx] } {str(annotation(answer, "ANSWER", "#8ef"))} {context[end_idx:]} \n '), # unsafe_allow_html=True, # ) st.write( markdown(f"Answer: {result}"), unsafe_allow_html=True, )