import logging from typing import List from pydantic import NoneStr import os from langchain.chains.question_answering import load_qa_chain from langchain.document_loaders import UnstructuredFileLoader from langchain.embeddings.openai import OpenAIEmbeddings from langchain.llms import OpenAI from langchain.text_splitter import CharacterTextSplitter from langchain.vectorstores import FAISS import gradio as gr import openai from langchain import PromptTemplate, OpenAI, LLMChain import validators import requests import mimetypes import tempfile import pandas as pd import re class DocumentQA: def get_empty_state(self): """ Create empty Knowledge base""" return {"knowledge_base": None} def get_content_from_url(self,url:str)->List: """ Uploads a file from a given URL and returns the loaded document. Args: url (str): The URL of the file to be uploaded. Returns: Document: The loaded document. Raises: ValueError: If the URL is not valid or the file cannot be fetched. """ if validators.url(url): headers = {'user-agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/114.0.0.0 Safari/537.36',} r = requests.get(url,headers=headers) if r.status_code != 200: raise ValueError( "Check the url of your file; returned status code %s" % r.status_code ) content_type = r.headers.get("content-type") file_extension = mimetypes.guess_extension(content_type) temp_file = tempfile.NamedTemporaryFile(suffix=file_extension, delete=False) temp_file.write(r.content) file_path = temp_file.name loader = UnstructuredFileLoader(file_path, strategy="fast") docs = loader.load() return docs else: raise ValueError("Please enter a valid URL") def create_knowledge_base(self,docs): """Create a knowledge base from the given documents. Args: docs (List[str]): List of documents. Returns: FAISS: Knowledge base built from the documents. """ # Initialize a CharacterTextSplitter to split the documents into chunks # Each chunk has a maximum length of 500 characters # There is no overlap between the chunks text_splitter = CharacterTextSplitter( separator="\n", chunk_size=500, chunk_overlap=100, length_function=len ) # Split the documents into chunks using the text_splitter chunks = text_splitter.split_documents(docs) # Initialize an OpenAIEmbeddings model to compute embeddings of the chunks embeddings = OpenAIEmbeddings() # Build a knowledge base using FAISS from the chunks and their embeddings knowledge_base = FAISS.from_documents(chunks, embeddings) # Return the resulting knowledge base return knowledge_base def get_chemicals_for_url(self,urls:str,state,input_qus)->str: """ Retrieves chemicals from the provided URLs. Args: urls (str): Comma-separated URLs of the files to be processed. Returns: str: The extracted chemical names. Raises: ValueError: If an error occurs during the process. """ webpage_text =[] for url in urls.split(','): webpage_text.extend(self.get_content_from_url(url)) knowledge_base = self.create_knowledge_base(webpage_text) state = {"knowledge_base": knowledge_base} chemicals = self.get_chemicals_for_file(state,input_qus) return chemicals def file_path_show(self,file_paths): file_paths = [single_file_path.name for single_file_path in file_paths] return file_paths def get_chemicals_for_file(self,state,question): knowledge_base = state["knowledge_base"] # Set the question for which we want to find the answer # question = "Identify the Chemical Capabilities Only" # Perform a similarity search on the knowledge base to retrieve relevant documents docs = knowledge_base.similarity_search(question) # Initialize an OpenAI language model for question answering template = """Use the following pieces of context to answer the question at the end. If you don't know the answer, just say that you don't know, don't try to make up an answer. Identify the Chemical Capabilities Only. {context} Question :{question}. The result should be in bullet points format. """ prompt = PromptTemplate(template=template,input_variables=["context","question"]) llm = OpenAI(temperature=0.4) llm_chain = LLMChain(prompt=prompt, llm=llm) # Load a question-answering chain using the language model chain = load_qa_chain(llm, chain_type="stuff",prompt=prompt) # Run the question-answering chain on the input documents and question response = chain.run(input_documents=docs, question=question) # Return the response as the answer to the question return response def identify_chemicals_in_files(self,file_paths,state,question): """Upload a file and create a knowledge base from its contents. Args: file_paths : The files to uploaded. Returns: tuple: A tuple containing the file name and the knowledge base. """ file_paths = [single_file_path.name for single_file_path in file_paths] docs =[] for file_obj in file_paths: loader = UnstructuredFileLoader(file_obj, strategy="fast") # Load the contents of the file using the loader docs.extend(loader.load()) # Create a knowledge base from the loaded documents using the create_knowledge_base() method knowledge_base = self.create_knowledge_base(docs) state = {"knowledge_base": knowledge_base} pdf_name = os.path.basename(file_obj) final_ans = self.get_chemicals_for_file(state,question) # Return a tuple containing the file name and the knowledge base return final_ans def get_final_result(self,urls,file_paths,state,input_qus): if urls: if file_paths: urls_chemicals = self.get_chemicals_for_url(urls,state,input_qus) file_chemicals = self.identify_chemicals_in_files(file_paths,state,input_qus) chemicals = urls_chemicals + file_chemicals return chemicals else: urls_chemicals = self.get_chemicals_for_url(urls,state,input_qus) return urls_chemicals elif file_paths: file_chemicals = self.identify_chemicals_in_files(file_paths,state,input_qus) return file_chemicals else: return "No Files Uploaded" document_qa = DocumentQA() class ChemicalIdentifier: def __init__(self): openai.api_key = os.getenv("OPENAI_API_KEY") # os.environ['OPENAI_API_KEY'] = openai_api_key def get_empty_state(self): """ Create empty Knowledge base""" return {"knowledge_base": None} def get_content_from_url(self,url:str)->List: """ Uploads a file from a given URL and returns the loaded document. Args: url (str): The URL of the file to be uploaded. Returns: Document: The loaded document. Raises: ValueError: If the URL is not valid or the file cannot be fetched. """ if validators.url(url): headers = {'user-agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/114.0.0.0 Safari/537.36',} r = requests.get(url,headers=headers) if r.status_code != 200: raise ValueError( "Check the url of your file; returned status code %s" % r.status_code ) content_type = r.headers.get("content-type") file_extension = mimetypes.guess_extension(content_type) temp_file = tempfile.NamedTemporaryFile(suffix=file_extension, delete=False) temp_file.write(r.content) file_path = temp_file.name loader = UnstructuredFileLoader(file_path, strategy="fast") docs = loader.load() return docs else: raise ValueError("Please enter a valid URL") def create_knowledge_base(self,docs): """Create a knowledge base from the given documents. Args: docs (List[str]): List of documents. Returns: FAISS: Knowledge base built from the documents. """ # Initialize a CharacterTextSplitter to split the documents into chunks # Each chunk has a maximum length of 500 characters # There is no overlap between the chunks text_splitter = CharacterTextSplitter( separator="\n", chunk_size=1000, chunk_overlap=200, length_function=len ) # Split the documents into chunks using the text_splitter chunks = text_splitter.split_documents(docs) # Initialize an OpenAIEmbeddings model to compute embeddings of the chunks embeddings = OpenAIEmbeddings() # Build a knowledge base using FAISS from the chunks and their embeddings knowledge_base = FAISS.from_documents(chunks, embeddings) # Return the resulting knowledge base return knowledge_base def get_chemicals_for_url(self,urls:str,state)->str: """ Retrieves chemicals from the provided URLs. Args: urls (str): Comma-separated URLs of the files to be processed. Returns: str: The extracted chemical names. Raises: ValueError: If an error occurs during the process. """ total_chemical=[] for url in urls.split(','): webpage_text = self.get_content_from_url(url) knowledge_base = self.create_knowledge_base(webpage_text) state = {"knowledge_base": knowledge_base} chemicals = self.get_chemicals_for_file(state) total_chemical.append(str(url)+"\n"+chemicals+"\n\n") list_of_chemicals = "".join(total_chemical) return list_of_chemicals def file_path_show(self,file_paths): file_paths = [single_file_path.name for single_file_path in file_paths] return file_paths def get_chemicals_for_file(self,state): knowledge_base = state["knowledge_base"] # Set the question for which we want to find the answer question = "list out chemicals.Result should be in bullet form" # Perform a similarity search on the knowledge base to retrieve relevant documents docs = knowledge_base.similarity_search(question) # Initialize an OpenAI language model for question answering # template = """Use the following pieces of context to answer the question at the end. If you don't know the answer, just say that you don't know, don't try to make up an answer. # list out all the chemical names. # {context} # Question :{question}. # The result should be in bullet points format. # """ # prompt = PromptTemplate(template=template,input_variables=["context","question"]) llm = OpenAI(temperature=0.4) # Load a question-answering chain using the language model chain = load_qa_chain(llm, chain_type="stuff") # Run the question-answering chain on the input documents and question response = chain.run(input_documents=docs, question=question) # Return the response as the answer to the question return response def identify_chemicals_in_files(self,file_paths,state): """Upload a file and create a knowledge base from its contents. Args: file_paths : The files to uploaded. Returns: tuple: A tuple containing the file name and the knowledge base. """ file_paths = [single_file_path.name for single_file_path in file_paths] results ='' for file_obj in file_paths: loader = UnstructuredFileLoader(file_obj, strategy="fast") # Load the contents of the file using the loader docs =loader.load() # Create a knowledge base from the loaded documents using the create_knowledge_base() method knowledge_base = self.create_knowledge_base(docs) state = {"knowledge_base": knowledge_base} pdf_name = os.path.basename(file_obj) final_ans = self.get_chemicals_for_file(state) results += pdf_name+"\n"+final_ans+"\n\n" # Return a tuple containing the file name and the knowledge base return results def get_final_result(self,urls,file_paths,state): if urls: if file_paths: urls_chemicals = self.get_chemicals_for_url(urls,state) file_chemicals = self.identify_chemicals_in_files(file_paths,state) chemicals = urls_chemicals + file_chemicals return chemicals else: urls_chemicals = self.get_chemicals_for_url(urls,state) return urls_chemicals elif file_paths: file_chemicals = self.identify_chemicals_in_files(file_paths,state) return file_chemicals else: return "No Files Uploaded" def gradio_interface(self)->None: """ Starts the Gradio interface for chemical identification. """ with gr.Blocks(css="style.css",theme='karthikeyan-adople/hudsonhayes-gray') as demo: gr.HTML("""

-

-

Chemical Capability Identifier

""") state = gr.State(self.get_empty_state()) with gr.Column(elem_id="col-container"): with gr.Row(elem_id="row-flex"): url = gr.Textbox(label="URL") with gr.Row(elem_id="row-flex"): with gr.Accordion("Upload Files", open = False): with gr.Row(): with gr.Column(scale=0.90, min_width=160): file_output = gr.File() with gr.Column(scale=0.10, min_width=160): upload_button = gr.UploadButton( "Browse File", file_types=[".txt", ".pdf", ".doc", ".docx"], file_count = "multiple",variant="primary") with gr.Row(): with gr.Column(scale=1, min_width=0): compare_btn = gr.Button(value="Generate Analysis",variant="primary") with gr.Row(): with gr.Column(scale=1, min_width=0): compared_result = gr.Textbox(value="",label='Chemical Capabilities :',show_label=True, placeholder="",lines=10) with gr.Row(): with gr.Column(scale=1, min_width=0): input_qus = gr.Textbox(value="",label='Question :',show_label=True, placeholder="") with gr.Row(): with gr.Column(scale=1, min_width=0): find_answer = gr.Button(value="Find Answer",label='Find',show_label=True, placeholder="") with gr.Row(): with gr.Column(scale=1, min_width=0): output = gr.Textbox(value="",label='Answer:',show_label=True, placeholder="") upload_button.upload(self.file_path_show, upload_button, [file_output]) compare_btn.click(self.get_final_result,[url,upload_button,state],compared_result) find_answer.click(document_qa.get_final_result,[url,upload_button,state,input_qus],output) demo.launch(debug=True) if __name__ == "__main__": chemical_identifier = ChemicalIdentifier() chemical_identifier.gradio_interface()