app.py

import os
import json
import streamlit as st
from transformers import AutoTokenizer, AutoModelForCausalLM, Trainer, TrainingArguments
from transformers import DataCollatorForLanguageModeling
import torch
from huggingface_hub import Repository, HfFolder
import subprocess

# Authenticate Hugging Face Hub
hf_token = st.secrets["HF_TOKEN"]  # Store your token in the Hugging Face Space Secrets
HfFolder.save_token(hf_token)

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

set_git_config()

@st.cache_data
def load_data(file_path):
    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)
        return data
    except Exception as e:
        st.error(f"Error loading dataset: {str(e)}")
        return None

@st.cache_resource
def initialize_model_and_tokenizer(model_name):
    try:
        tokenizer = AutoTokenizer.from_pretrained(model_name)
        model = AutoModelForCausalLM.from_pretrained(model_name)
        
        # Set the pad token to the eos token if it doesn't exist
        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):
    inputs = []
    for item in data:
        prompt = item['prompt']
        response = item['response']
        full_text = f"Human: {prompt}\nAssistant: {response}"
        encoded = tokenizer.encode_plus(
            full_text,
            max_length=max_length,
            padding='max_length',
            truncation=True,
            return_tensors='pt'
        )
        inputs.append({
            'input_ids': encoded['input_ids'].squeeze(),
            'attention_mask': encoded['attention_mask'].squeeze()
        })
    return inputs

class SimpleDataset(torch.utils.data.Dataset):
    def __init__(self, encodings):
        self.encodings = encodings

    def __getitem__(self, idx):
        return {key: val[idx] for key, val in self.encodings[idx].items()}

    def __len__(self):
        return len(self.encodings)

def main():
    st.title("Model Training with Streamlit")

    # User inputs with recommended values
    model_name = st.text_input("Enter model name", "distilgpt2")
    file_path = st.text_input("Enter path to training data JSON file", "appointment_training_data.json")
    max_length = st.number_input("Enter max token length", min_value=32, max_value=512, value=256)
    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")
    
    # Specify your Hugging Face model repository ID
    repo_id = st.text_input("Enter Hugging Face repository ID", "nileshhanotia/PeVe")

    tokenizer, model = initialize_model_and_tokenizer(model_name)
    
    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_path)
    
    if data is None:
        st.warning("Failed to load dataset. Please check the file path and try again.")
        return

    st.write("Tokenizing dataset...")
    tokenized_dataset = create_dataset(data, tokenizer, max_length)

    dataset = SimpleDataset(tokenized_dataset)

    # Define training arguments
    training_args = TrainingArguments(
        output_dir='./results',
        evaluation_strategy='no',
        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,
    )

    # Initialize the Trainer
    trainer = Trainer(
        model=model,
        args=training_args,
        train_dataset=dataset,
        data_collator=DataCollatorForLanguageModeling(tokenizer=tokenizer, mlm=False),
    )

    if st.button('Start Training'):
        st.write("Starting training...")
        progress_bar = st.progress(0)

        # Initialize the Hugging Face repository
        repo = Repository(local_dir="./results", clone_from=repo_id)
        
        for epoch in range(int(num_epochs)):
            trainer.train()
            progress = (epoch + 1) / num_epochs
            progress_bar.progress(progress)
            
            # Save the model after each epoch locally
            model_path = f"./results/model_epoch_{epoch+1}"
            trainer.save_model(model_path)
            st.write(f"Model saved locally: {model_path}")

            # Push to Hugging Face Hub
            repo.push_to_hub(commit_message=f"Model after epoch {epoch+1}")
            st.write(f"Model pushed to Hugging Face Hub: {repo_id}")

        st.write("Training complete. Model is available on the Hugging Face Hub.")

if __name__ == "__main__":
    main()