app.py

import pandas as pd
import chromadb
from sklearn.model_selection import train_test_split
from transformers import GPT2Tokenizer, GPT2LMHeadModel, TextDataset, DataCollatorForLanguageModeling, Trainer, TrainingArguments, pipeline
import gradio as gr
import email

# loading and preprocessing dataset
emails = pd.read_csv('emails.csv')

def preprocess_email_content(raw_email):
    message = email.message_from_string(raw_email).get_payload()
    return message.replace("\n", "").replace("\r", "").replace("> >>> > >", "").strip()

content_text = [preprocess_email_content(item) for item in emails['message']]
train_content, _ = train_test_split(content_text, train_size=0.00005)

# ChromaDB setup
client = chromadb.Client()
collection = client.create_collection(name="Enron_emails")
collection.add(documents=train_content, ids=[f'id{i+1}' for i in range(len(train_content))])

# initialize model and tokenizer globally but don't load them yet
tokenizer = None
model = None
text_gen = None

def load_model():
    global tokenizer, model, text_gen
    if model is None or tokenizer is None:
        tokenizer = GPT2Tokenizer.from_pretrained('./fine_tuned_model')
        model = GPT2LMHeadModel.from_pretrained('./fine_tuned_model')
        tokenizer.add_special_tokens({'pad_token': '[PAD]'})
        text_gen = pipeline("text-generation", model=model, tokenizer=tokenizer)

def question_answer(question):
    load_model()  # loading model on first use
    try:
        generated = text_gen(question, max_length=200, num_return_sequences=1)
        generated_text = generated[0]['generated_text'].replace(question, "").strip()
        return generated_text
    except Exception as e:
        return f"Error in generating response: {str(e)}"

iface = gr.Interface(
    fn=question_answer,
    inputs="text",
    outputs="text",
    title="Answering questions about the Enron case.",
    description="Ask a question about the Enron case!",
    examples=["What is Enron?"]
)
iface.launch()