app.py

import json
import logging
import os
import shutil
import sys
import uuid
from json import JSONDecodeError
from pathlib import Path
from tqdm.auto import tqdm
import datetime
from time import sleep
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)

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


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("<style>.stButton button {width:100%;}</style>", unsafe_allow_html=True)
col2.markdown("<style>.stButton button {width:100%;}</style>", 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(
            "🧠 &nbsp;&nbsp; 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("👓 &nbsp;&nbsp; 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("🧑‍🌾 &nbsp;&nbsp; All our workers are busy! Try again later.")
                else:
                    st.error(f"🐞 &nbsp;&nbsp; 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,
            )