import pickle import sys from argparse import ArgumentParser from pathlib import Path import dotenv import faiss import tiktoken from langchain_openai import OpenAIEmbeddings from langchain.text_splitter import CharacterTextSplitter from langchain.vectorstores import FAISS def num_tokens_from_string(string: str, encoding_name: str) -> int: # Function to convert string to tokens and estimate user cost. encoding = tiktoken.get_encoding(encoding_name) num_tokens = len(encoding.encode(string)) total_price = ((num_tokens / 1000) * 0.0004) return num_tokens, total_price def call_openai_api(): # Function to create a vector store from the documents and save it to disk. store = FAISS.from_texts(docs, OpenAIEmbeddings(), metadatas=metadatas) faiss.write_index(store.index, "docs.index") store.index = None with open("faiss_store.pkl", "wb") as f: pickle.dump(store, f) def get_user_permission(): # Function to ask user permission to call the OpenAI api and spend their OpenAI funds. # Here we convert the docs list to a string and calculate the number of OpenAI tokens the string represents. docs_content = (" ".join(docs)) tokens, total_price = num_tokens_from_string(string=docs_content, encoding_name="cl100k_base") # Here we print the number of tokens and the approx user cost with some visually appealing formatting. print(f"Number of Tokens = {format(tokens, ',d')}") print(f"Approx Cost = ${format(total_price, ',.2f')}") # Here we check for user permission before calling the API. user_input = input("Price Okay? (Y/N) \n").lower() if user_input == "y": call_openai_api() elif user_input == "": call_openai_api() else: print("The API was not called. No money was spent.") # Load .env file dotenv.load_dotenv() ap = ArgumentParser("Script for training DocsGPT on .rst documentation files.") ap.add_argument("-i", "--inputs", type=str, default="inputs", help="Directory containing documentation files") args = ap.parse_args() # Here we load in the data in the format that Notion exports it in. ps = list(Path(args.inputs).glob("**/*.rst")) # parse all child directories data = [] sources = [] for p in ps: with open(p) as f: data.append(f.read()) sources.append(p) # Here we split the documents, as needed, into smaller chunks. # We do this due to the context limits of the LLMs. text_splitter = CharacterTextSplitter(chunk_size=1500, separator="\n") docs = [] metadatas = [] for i, d in enumerate(data): splits = text_splitter.split_text(d) docs.extend(splits) metadatas.extend([{"source": sources[i]}] * len(splits)) # Here we check for command line arguments for bot calls. # If no argument exists or the permission_bypass_flag argument is not '-y', # user permission is requested to call the API. if len(sys.argv) > 1: permission_bypass_flag = sys.argv[1] if permission_bypass_flag == '-y': call_openai_api() else: get_user_permission() else: get_user_permission()