Optimizing model training with Lightning

By implementing automated techniques like ModelCheckpoint and EarlyStopping, you'll ensure your model selects the best-performing parameters while avoiding unnecessary computations.

The dataset, a subset of the Osmanya MNIST dataset, provides a real-world use case where scalable AI training techniques can significantly improve efficiency and accuracy.

OsmanyaDataModule and ImageClassifier have been predefined for you.

This exercise is part of the course

Scalable AI Models with PyTorch Lightning

Exercise instructions

Import callbacks that you'll use for model checkpointing and early stopping.
Train the model with the ModelCheckpoint and EarlyStopping callbacks.

Hands-on interactive exercise

Have a go at this exercise by completing this sample code.

# Import relevant checkpoints
from lightning.pytorch.callbacks import ____, ____

class EvaluatedImageClassifier(ImageClassifier):
    def validation_step(self, batch, batch_idx):
        x, y = batch
        y_hat = self(x)
        acc = (y_hat.argmax(dim=1) == y).float().mean()
        self.log("val_acc", acc)

model = EvaluatedImageClassifier()
data_module = OsmanyaDataModule()
# Train the model with ModelCheckpoint and EarlyStopping checkpoints
trainer = Trainer(____=[____(monitor="val_acc", save_top_k=1), ____(monitor="val_acc", patience=3)])
trainer.fit(model, datamodule=data_module)

Edit and Run Code

This exercise is part of the course

Scalable AI Models with PyTorch Lightning

IntermediateSkill Level

5.0+

Start Course for Free

In this chapter, we'll explore how PyTorch Lightning simplifies the development and deployment of scalable AI models. Starting with foundational concepts, we'll go through the core structure of a PyTorch Lightning project, including essential components like the LightningModule and Trainer, to set a strong foundation for more advanced AI solutions.

Exercise 1: Introduction to PyTorch Lightning Exercise 2: Introducing the LightningModule Exercise 3: Running the Lightning Trainer Exercise 4: Defining models with LightningModule Exercise 5: Usage of the LightningModule Exercise 6: Mastering the init method Exercise 7: Perfecting the forward method Exercise 8: Implementing training logic Exercise 9: Implementing the training step Exercise 10: Configuring the optimizer Exercise 11: Training and evaluating

We'll dive deeper into PyTorch Lightning to efficiently manage data and refine model training in this chapter. We'll learn how to create modular and reusable data workflows with LightningDataModule, evaluate your models accurately through validation and testing, and enhance training processes using Lightning Callbacks to automate model improvement and avoid overfitting.

Exercise 1: Managing data with LightningDataModule Exercise 2: Splitting data with LightningDataModule Exercise 3: Creating a train DataLoader Exercise 4: Incorporating validation and testing Exercise 5: Implementing the validation step Exercise 6: Evaluate model accuracy using Torchmetrics Exercise 7: Enhancing training with Lightning callbacks Exercise 8: Classifying Lightning callbacks Exercise 9: Optimizing model training with Lightning

Current Exercise

Learn to prepare deep learning models for real-world deployment by making them leaner and faster. This chapter introduces techniques such as dynamic quantization, pruning, and TorchScript conversion, helping you reduce model size and latency without sacrificing accuracy

Exercise 1: Applying dynamic quantization Exercise 2: Apply dynamic quantization Exercise 3: Comparing quantized model performance Exercise 4: Implementing model pruning techniques Exercise 5: Apply pruning to a linear layer Exercise 6: Finalize pruning by removing the mask Exercise 7: Exporting models with TorchScript Exercise 8: Choosing the right conversion method Exercise 9: Optimizing models for scalability Exercise 10: Recap: Scalable AI Models with PyTorch Lightning