Common Workflows

Customize and extend Lightning for things like custom hardware or distributed strategies.

Avoid overfitting

Add a training and test loop.

Build a model

Steps to build a model.

Configure hyperparameters from the CLI

Enable basic CLI with Lightning.

Customize the progress bar

Change the progress bar behavior.

Deploy models into production

Deploy models with different levels of scale.

Effective Training Techniques

Explore advanced training techniques.

Eliminate config boilerplate

Control your training via CLI and YAML.

Find bottlenecks in your code

Learn to find bottlenecks in your code.

Finetune a model

Learn to use pretrained models

Manage Experiments

Learn to track and visualize experiments

Run on an on-prem cluster

Learn to run on your own cluster

Save and load model progress

Save and load progress with checkpoints.

Save memory with half-precision

Enable half-precision to train faster and save memory.

Train 1 trillion+ parameter models

Scale GPU training to 1 trillion + parameter models

Train on multiple machines over the internet

Train on local machines or unreliable GPUs across the internet.

Train on single or multiple GPUs

Train models faster with GPUs.

Train on single or multiple HPUs

Train models faster with HPUs.

Train on single or multiple IPUs

Train models faster with IPUs.

Train on single or multiple TPUs

Train models faster with TPUs.

Track and Visualize Experiments

Learn to track and visualize experiments

Use a pure PyTorch training loop

Run your pure PyTorch loop with Lightning.