app.py

import pandas as pd
import numpy as np
import gradio as gr
from transformers import GPT2LMHeadModel, GPT2Tokenizer, AutoTokenizer, AutoModel
import torch
from sentence_transformers import SentenceTransformer, util
from tqdm import tqdm
from peewee import SqliteDatabase, Model, TextField

# Load the Enron Email Dataset
emails_df = pd.read_csv("/content/emails.csv")

# Define the ChromaDB database
db = SqliteDatabase('email_embeddings.db')

# Define the model and tokenizer
tokenizer = GPT2Tokenizer.from_pretrained("gpt2")
model = GPT2LMHeadModel.from_pretrained("gpt2")

# Fine-tune the Language Model on the dataset
# Fine-tune the Language Model on the dataset
# Tokenize the dataset
tokenized_texts = [tokenizer.encode(text, return_tensors="pt", max_length=512) for text in emails_df['Message']]

# Convert tokenized texts to tensors
input_ids = torch.cat(tokenized_texts, dim=1)

# Define a PyTorch dataset
dataset = torch.utils.data.TensorDataset(input_ids)


# Define the Sentence Transformer model
sentence_model = SentenceTransformer('distilbert-base-nli-mean-tokens')

# Function to create embeddings of the email dataset and store them in the ChromaDB database
def create_embeddings():
    db.connect()
    db.create_tables([Email])

    embeddings = []

    for index, row in tqdm(emails_df.iterrows(), total=len(emails_df)):
        text = row['Message']
        embeddings.append(sentence_model.encode(text))

    for index, embedding in enumerate(embeddings):
        Email.create(id=index, embedding=embedding.tobytes())

    db.close()

# Define the Gradio Interface
def answer_question(question):
    # Encode the question
    inputs = tokenizer(question, return_tensors="pt", max_length=512, truncation=True)

    # Generate response using the model
    outputs = model.generate(**inputs)

    # Decode the response
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)

    return response

# Define the Peewee Model for the ChromaDB database
class BaseModel(Model):
    class Meta:
        database = db

class Email(BaseModel):
    embedding = TextField()

# Create a Gradio Interface
gr.Interface(fn=answer_question, inputs="text", outputs="text").launch()

# Uncomment the line below to create embeddings and store them in the ChromaDB database
create_embeddings()