from flask import Flask, request import os import requests from langchain.vectorstores import Chroma from langchain.llms import OpenAI from langchain.chains import RetrievalQA from InstructorEmbedding import INSTRUCTOR from langchain.embeddings import HuggingFaceInstructEmbeddings from langchain.chat_models import ChatOpenAI import numpy import torch import json import textwrap from flask_cors import CORS import socket; import gradio as gr app = Flask(__name__) cors = CORS(app) def get_local_ip(): s = socket.socket(socket.AF_INET, socket.SOCK_DGRAM) s.connect(("8.8.8.8", 80)) return s.getsockname()[0] def wrap_text_preserve_newlines(text, width=110): # Split the input text into lines based on newline characters lines = text.split('\n') # Wrap each line individually wrapped_lines = [textwrap.fill(line, width=width) for line in lines] # Join the wrapped lines back together using newline characters wrapped_text = '\n'.join(wrapped_lines) return wrapped_text def process_llm_response(llm_response): response_data = { 'result': wrap_text_preserve_newlines(llm_response['result']), 'sources': [] } print(wrap_text_preserve_newlines(llm_response['result'])) print('\n\nSources:') for source in llm_response["source_documents"]: print(source.metadata['source']+ "Page Number: " + str(source.metadata['page'])) response_data['sources'].append({"book": source.metadata['source'], "page": source.metadata['page']}) # return json.dumps(response_data) return response_data # @app.route('/question', methods=['POST']) # def answer(): # content_type = request.headers.get('Content-Type') # if (content_type == 'application/json'): # data = request.json # question = data['question'] # response = get_answer(question) # return response # else: # return 'Content-Type not supported!' ip=get_local_ip() os.environ["OPENAI_API_KEY"] = "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" # Embed and store the texts # if(torch.cuda.is_available() == False): # print("No GPU available") # exit(1) torch.cuda.empty_cache() torch.max_split_size_mb = 100 instructor_embeddings = HuggingFaceInstructEmbeddings(model_name="hkunlp/instructor-xl", model_kwargs={"device": "cpu"}) # Supplying a persist_directory will store the embeddings on disk persist_directory = 'db' vectordb2 = Chroma(persist_directory=persist_directory, embedding_function=instructor_embeddings, ) retriever = vectordb2.as_retriever(search_kwargs={"k": 3}) vectordb2.persist() # Set up the turbo LLM turbo_llm = ChatOpenAI( temperature=0, model_name='gpt-3.5-turbo' ) qa_chain = RetrievalQA.from_chain_type(llm=turbo_llm, chain_type="stuff", retriever=retriever, return_source_documents=True) qa_chain.combine_documents_chain.llm_chain.prompt.messages[0].prompt.template= """ Use only the following pieces of context. Answer the users question only if they are related to the context given. If you don't know the answer, just say that you don't know, don't try to make up an answer. Make your answer very detailed and long. Use bullet points to explain when required. Use only text found in the context as your knowledge source for the answer. ---------------- {context}""" def book_url(book): if book == "BD Human Anatomy - Lower Limb, Abdomen & Pelvis (Volume 2).pdf": return "BD+Human+Anatomy+-+Lower+Limb%2C+Abdomen+%26+Pelvis+(Volume+2).pdf" elif book == "BD Human Anatomy - Upper Limb & Thorax (Volume 1).pdf": return "BD+Human+Anatomy+-+Upper+Limb++Thorax+(Volume+1).pdf" elif book == "[Richard S.Snell] Clinical Neuroanatomy (7th Ed.)": return "%5BRichard+S.Snell%5D+Clinical+Neuroanatomy+(7th+Ed.).pdf" elif book == "BD Chaurasia's Handbook of General Anatomy, 4th Edition.pdf": return "BD+Chaurasia's+Handbook+of+General+Anatomy%2C+4th+Edition.pdf" elif book == "Vishram Singh Textbook of Anatomy Upper Limb and Thorax..pdf": return "BD+Chaurasia's+Handbook+of+General+Anatomy%2C+4th+Edition.pdf" elif book == "Vishram Singh Textbook of Anatomy Vol 2.pdf": return "Vishram+Singh+Textbook+of+Anatomy+Vol+2.pdf" elif book == "BD Human Anatomy - Head, Neck & Brain (Volume 3).pdf": return "BD+Human+Anatomy+-+Head%2C+Neck+%26+Brain+(Volume+3).pdf" elif book == "Textbook of Clinical Neuroanatomy.pdf": return "Textbook+of+Clinical+Neuroanatomy.pdf" elif book == "Vishram Singh Textbook of Anatomy Vol 3.pdf": return "Vishram+Singh+Textbook+of+Anatomy+Vol+3.pdf" def print_array(arr): # Convert the array to a string representation arr_str = str(arr) return arr_str def html_link_generator(book, page): bookurl = book_url(book) url = f"https://diagrams1.s3.ap-south-1.amazonaws.com/anatomybooks/{bookurl}#page={page}" # html = f'' # print(url) return url def getanswer(question): if question=="" : return "Please ask a question" , [] llm_response = qa_chain(question) response = process_llm_response(llm_response) html= html_link_generator(response["sources"][0]["book"][22:], response["sources"][0]["page"]) # html = """""" return response["result"], response['sources'] def makevisible(source1,source2,source3): return{ source1: gr.update(visible=True), source2: gr.update(visible=True), source3: gr.update(visible=True) } with gr.Blocks() as demo: with gr.Row(): with gr.Column(scale=1, min_width=600): question = gr.Textbox(label="Question") submitbtn = gr.Button("Submit").style(full_width=True) answer = gr.Textbox(label="Answer", interactive=False) sources = gr.Json(label="Sources", interactive=False) source1 = gr.Button(label="Source 1", visible=False) source2 = gr.Button(label="Source 2", visible=False) source3 = gr.Button(label="Source 3", visible=False) submitbtn.click(fn=getanswer, inputs=[question], outputs=[answer, sources], api_name="question") # source1.click(fn=None, _js=f"""window.open('"""+"""', target="_blank");""") # sources.change(make_source_buttons, [sources, source1, source2, source3], [source1,source2,source3]) demo.launch()