app.py

import os
import json
import random
import streamlit as st
from transformers import AutoTokenizer, AutoModelForSequenceClassification, Trainer, TrainingArguments
from datasets import Dataset, concatenate_datasets
import torch
from huggingface_hub import Repository, HfFolder
import subprocess

# Authenticate Hugging Face Hub
hf_token = st.secrets["HF_TOKEN"]
HfFolder.save_token(hf_token)

# Set Git user identity
def set_git_config():
    try:
        subprocess.run(['git', 'config', '--global', 'user.email', '[email protected]'], check=True)
        subprocess.run(['git', 'config', '--global', 'user.name', 'Nilesh'], check=True)
        st.success("Git configuration set successfully.")
    except subprocess.CalledProcessError as e:
        st.error(f"Git configuration error: {str(e)}")

# Call set_git_config at the start of the script
set_git_config()

@st.cache_data
def load_data(file_paths):
    combined_data = []
    for file_path in file_paths:
        file_path = file_path.strip()
        if not os.path.exists(file_path):
            st.error(f"File not found: {file_path}")
            return None
        try:
            with open(file_path, 'r') as f:
                data = json.load(f)
            if 'intents' in data:
                for intent in data['intents']:
                    combined_data.extend(intent['examples'])
            else:
                st.error(f"Invalid format in file: {file_path}")
                return None
        except Exception as e:
            st.error(f"Error loading dataset from {file_path}: {str(e)}")
            return None
    return combined_data

@st.cache_resource
def initialize_model_and_tokenizer(model_name, num_labels):
    try:
        tokenizer = AutoTokenizer.from_pretrained(model_name)
        model = AutoModelForSequenceClassification.from_pretrained(model_name, num_labels=num_labels)
        
        if tokenizer.pad_token is None:
            tokenizer.pad_token = tokenizer.eos_token
            model.config.pad_token_id = model.config.eos_token_id
        
        return tokenizer, model
    except Exception as e:
        st.error(f"Error initializing model and tokenizer: {str(e)}")
        return None, None

def create_dataset(data, tokenizer, max_length):
    texts = [item.get('prompt', '') for item in data]
    labels = [item.get('label', -1) for item in data]
    
    # Debugging: Print out labels to check for invalid values
    print(f"Labels before adjustment: {labels}")
    
    # Ensure all labels are within the valid range
    labels = [label if 0 <= label < num_labels else 0 for label in labels]
    
    # Debugging: Print out adjusted labels
    print(f"Labels after adjustment: {labels}")
    
    encodings = tokenizer(texts, truncation=True, padding='max_length', max_length=max_length)
    dataset = Dataset.from_dict({
        'input_ids': encodings['input_ids'],
        'attention_mask': encodings['attention_mask'],
        'labels': labels
    })
    return dataset


def split_data(data, test_size=0.2):
    if not data:
        raise ValueError("Data is empty, cannot split.")
    random.shuffle(data)
    split_index = int(len(data) * (1 - test_size))
    return data[:split_index], data[split_index:]

def main():
    st.title("Appointment Classification Model Training")

    model_name = st.text_input("Enter model name", "distilgpt2")
    file_paths = st.text_area("Enter paths to training data JSON files (comma-separated)", "training_data1.json,training_data2.json").split(',')
    max_length = st.number_input("Enter max token length", min_value=32, max_value=512, value=128)
    num_epochs = st.number_input("Enter number of training epochs", min_value=1, max_value=10, value=3)
    batch_size = st.number_input("Enter batch size", min_value=1, max_value=32, value=8)
    learning_rate = st.number_input("Enter learning rate", min_value=1e-6, max_value=1e-3, value=5e-5, format="%.1e")
    num_labels = 3  # We have 3 classes: schedule, reschedule, cancel
    
    repo_id = st.text_input("Enter Hugging Face repository ID", "nileshhanotia/PeVe")

    tokenizer, model = initialize_model_and_tokenizer(model_name, num_labels)
    
    if tokenizer is None or model is None:
        st.warning("Failed to initialize model and tokenizer. Please check the model name and try again.")
        return

    st.write("Loading and processing dataset...")
    data = load_data(file_paths)
    
    if data is None:
        st.warning("Failed to load dataset. Please check the file paths and try again.")
        return

    st.write("Preparing dataset...")
    # Split the data into train and evaluation sets
    try:
        train_data, eval_data = split_data(data)
    except ValueError as e:
        st.error(f"Data splitting error: {str(e)}")
        return
    
    train_dataset = create_dataset(train_data, tokenizer, max_length)
    eval_dataset = create_dataset(eval_data, tokenizer, max_length)

    training_args = TrainingArguments(
        output_dir='./results',
        evaluation_strategy='epoch',
        learning_rate=learning_rate,
        per_device_train_batch_size=batch_size,
        per_device_eval_batch_size=batch_size,
        num_train_epochs=num_epochs,
        weight_decay=0.01,
        logging_dir='./logs',
        logging_steps=10,
        push_to_hub=True,
        hub_model_id=repo_id,
    )

    trainer = Trainer(
        model=model,
        args=training_args,
        train_dataset=train_dataset,
        eval_dataset=eval_dataset,
    )

    if st.button('Start Training'):
        st.write("Starting training...")
        trainer.train()
        trainer.push_to_hub()
        st.write(f"Training complete. Model is available on the Hugging Face Hub: {repo_id}")

if __name__ == "__main__":
    main()