app.py

import gradio as gr
import torch
from datasets import load_dataset
from transformers import (
    AutoTokenizer,
    AutoModelForSequenceClassification,
    TrainingArguments,
    Trainer
)

def train_cpu_optimized():
    """Train TinyBERT with CPU Basic optimized settings"""
    model_name = "huawei-noah/TinyBERT_General_4L_312D"
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModelForSequenceClassification.from_pretrained(model_name, num_labels=2)

    # Load IMDB dataset
    raw_dataset = load_dataset("imdb")

    def tokenize_function(examples):
        return tokenizer(examples["text"], padding="max_length", truncation=True, max_length=64)

    # CPU-friendly dataset sizes
    train_ds = raw_dataset["train"].shuffle(seed=42).select(range(500))
    eval_ds = raw_dataset["test"].shuffle(seed=42).select(range(200))

    train_dataset = train_ds.map(tokenize_function, batched=True)
    eval_dataset = eval_ds.map(tokenize_function, batched=True)

    # CPU-optimized training arguments
    training_args = TrainingArguments(
        output_dir="./results",
        num_train_epochs=3,
        per_device_train_batch_size=8,
        per_device_eval_batch_size=16,
        learning_rate=3e-4,
        evaluation_strategy="epoch",
        save_strategy="epoch",
        logging_steps=25,
        dataloader_num_workers=0,
        report_to="none",
    )

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

    # Start training
    trainer.train()

    # Save the final model
    trainer.save_model("./final_tinybert_model")
    tokenizer.save_pretrained("./final_tinybert_model")

    return "✅ Training complete! Model saved to ./final_tinybert_model"

def test_model(text):
    """Test your trained model"""
    try:
        from transformers import pipeline
        pipe = pipeline("sentiment-analysis", model="./final_tinybert_model")
        result = pipe(text)
        return f"Prediction: {result[0]['label']} (Confidence: {result[0]['score']:.3f})"
    except Exception as e:
        return f"Error: {str(e)}. Please train the model first!"

# Create Gradio interface
with gr.Blocks() as demo:
    gr.Markdown("# 🤖 TinyBERT CPU-Optimized Training")
    gr.Markdown("**Complete ML workflow on CPU Basic - perfectly optimized for your hardware!**")

    with gr.Tab("🚀 Train Model"):
        gr.Markdown("This will train TinyBERT on 500 IMDB samples (15-20 minutes)")
        train_btn = gr.Button("Start CPU-Optimized Training")
        train_output = gr.Textbox(label="Training Progress", lines=5)
        train_btn.click(train_cpu_optimized, outputs=train_output)

    with gr.Tab("🧪 Test Model"):
        gr.Markdown("Test your trained sentiment analysis model:")
        test_input = gr.Textbox(label="Enter text to analyze", placeholder="This movie was fantastic!")
        test_btn = gr.Button("Analyze Sentiment")
        test_output = gr.Textbox(label="Prediction Result")
        test_btn.click(test_model, inputs=test_input, outputs=test_output)

demo.launch()