Gradient checkpointing with Accelerator

You're continuing to optimize memory usage so you can train your language translation model on your device. Gradient accumulation has helped you to effectively train on larger batch sizes. Build on this work to add gradient checkpointing to reduce the memory footprint of your model.

The model, train_dataloader, and accelerator have been pre-defined.

This exercise is part of the course

Efficient AI Model Training with PyTorch

View Course

Exercise instructions

Enable gradient checkpointing on the model.
Set up an Accelerator context manager to enable gradient accumulation on the model.

Hands-on interactive exercise

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

# Enable gradient checkpointing on the model
____.____()

for batch in train_dataloader:
    with accelerator.accumulate(model):
        inputs, targets = batch["input_ids"], batch["labels"]
        # Get the outputs from a forward pass of the model
        ____ = ____(____, labels=targets)
        loss = outputs.loss
        accelerator.backward(loss)
        optimizer.step()
        lr_scheduler.step()
        optimizer.zero_grad()
        print(f"Loss = {loss}")

Edit and Run Code