app.py

import gradio as gr
from transformers import AutoTokenizer, AutoModelForCausalLM, Trainer, TrainingArguments
from datasets import load_dataset

# Define datasets and their IDs
datasets_info = {
    "SQuAD": "squad",
    "SQuAD 2.0": "squad_v2",
    "Natural Questions": "nq",
    "TriviaQA": "triviaqa",
    "QuAC": "quac",
    "FAQ Dataset": "faq",
    "BoolQ": "boolq",
    "Open Book QA": "obqa"
}

# Load model and tokenizer directly
tokenizer = AutoTokenizer.from_pretrained("nvidia/Llama-3.1-Nemotron-70B-Instruct-HF")
model = AutoModelForCausalLM.from_pretrained("nvidia/Llama-3.1-Nemotron-70B-Instruct-HF")

def train_model(dataset_name):
    # Load the dataset
    dataset = load_dataset(datasets_info[dataset_name])

    # Tokenization
    def preprocess_function(examples):
        return tokenizer(examples['question'], examples['context'], truncation=True)

    tokenized_dataset = dataset.map(preprocess_function, batched=True)

    # Fine-tune the model
    training_args = TrainingArguments(
        output_dir=f"./{dataset_name}_model",
        evaluation_strategy="epoch",
        learning_rate=2e-5,
        per_device_train_batch_size=8,
        per_device_eval_batch_size=8,
        num_train_epochs=3,
        weight_decay=0.01,
        logging_dir='./logs',
    )

    trainer = Trainer(
        model=model,
        args=training_args,
        train_dataset=tokenized_dataset['train'],
        eval_dataset=tokenized_dataset['validation']
    )

    trainer.train()
    
    # Save the model weights
    model.save_pretrained(f"./{dataset_name}_model")
    tokenizer.save_pretrained(f"./{dataset_name}_model")
    
    return f"Model trained and saved for {dataset_name}!"

# Gradio Interface
with gr.Blocks() as demo:
    gr.Markdown("## Train QA Model on Multiple Datasets")
    dataset_name = gr.Dropdown(choices=list(datasets_info.keys()), label="Select Dataset")
    train_button = gr.Button("Train Model")
    output = gr.Textbox(label="Output")

    def train_and_display(dataset_name):
        return train_model(dataset_name)

    train_button.click(train_and_display, inputs=dataset_name, outputs=output)

demo.launch()