app.py

from gpt_index import GPTSimpleVectorIndex
from langchain import OpenAI
import gradio as gr
from gradio import Interface, Textbox
import os
import datetime
from datasets import load_dataset
from huggingface_hub import HfFolder

os.environ["OPENAI_API_KEY"] = os.environ['SECRET_CODE']

# Need to write to persistent dataset because cannot store temp data on spaces
DATASET_REPO_URL = "https://huggingface.co/datasets/peterpull/MediatorBot"
DATA_FILENAME = "data.txt"
INDEX_FILENAME = "index_base_89MB.json"
DATA_FILE = os.path.join("data", DATA_FILENAME)
INDEX_FILE = os.path.join("data", INDEX_FILENAME)

# we need a write access token.
HF_TOKEN = os.environ.get("HF_TOKEN")
print("HF TOKEN is none?", HF_TOKEN is None)

# Clones the distant repo to the local repo
dataset = load_dataset(DATASET_REPO_URL)
dataset_folder = HfFolder(dataset._data_files["train"][0].path).path
print(f"Dataset folder: {dataset_folder}")
print(f"Dataset files: {os.listdir(dataset_folder)}")

def generate_text() -> str:
    with open(os.path.join(dataset_folder, DATA_FILENAME)) as file:
        text = ""
        for line in file:
            row_parts = line.strip().split(",")
            if len(row_parts) != 3:
                continue
            user, chatbot, time = row_parts
            text += f"Time: {time}\nUser: {user}\nChatbot: {chatbot}\n\n"
        return text if text else "No messages yet"


def store_message(chatinput: str, chatresponse: str):
    if chatinput and chatresponse:
        with open(os.path.join(dataset_folder, DATA_FILENAME), "a") as file:
            file.write(f"{datetime.datetime.now()},{chatinput},{chatresponse}\n")
            print(f"Wrote to datafile: {datetime.datetime.now()},{chatinput},{chatresponse}\n")

        # Push back to hub every N-th time the function is called
        if store_message.count_calls % 1 == 0:
            print("Pushing back to Hugging Face model hub")
            dataset.commit("Added new chat data")  # Commit the changes
        store_message.count_calls += 1
        
        return generate_text()

store_message.count_calls = 1  #initiases the count at one. We want to count how many messages stored before pushing back to repo.

# gets the index file which is the context data
def get_index(index_file_path):
    if os.path.exists(index_file_path):
        index_size = os.path.getsize(index_file_path)
        print(f"Size of {index_file_path}: {index_size} bytes") #let me know how big json file is.
        return GPTSimpleVectorIndex.load_from_disk(index_file_path)
    else:
        print(f"Error: '{index_file_path}' does not exist.")
        sys.exit()

index = get_index(INDEX_FILE)

# passes the prompt to the chatbot, queries the index, stores the output, returns the response
def chatbot(input_text, mentioned_person='Mediator John Haynes', confidence_threshold=0.5):
    prompt = f"You are {mentioned_person}. Answer this: {input_text}. Only reply from the contextual data, or say you don't know. At the end of your answer ask an insightful question."
    response = index.query(prompt, response_mode="default")

    store_message(input_text,response)
    
    # return the response
    return response.response


with open('about.txt', 'r') as file:
    about = file.read()

iface = Interface(
    fn=chatbot,
    inputs=Textbox("Enter your question"),
    outputs="text",
    title="AI Chatbot trained on J. Haynes mediation material, v0.5",
    description=about)
                                         
iface.launch()