Lightning & Multi Node Training

Lightning supports makes multi-node training simple by providing a simple interface to orchestrate compute and data.

Multi Node with raw PyTorch

You can run the multi-node raw PyTorch by running the following commands.

Here is an example where you spawn your processes yourself.

lightning run app train_pytorch.py

or you can use the built-in component for it.

lightning run app train_pytorch_spawn.py

You can run the multi-node raw PyTorch and Lite by running the following commands.

lightning run app train_lite.py

Using Lite, you retain control over your loops while accessing in a minimal way all Lightning distributed strategies.

Lightning supports running Lightning Trainer from a script or within a Lightning Work.

You can either run a script directly

lightning run app train_pl_script.py

or run your code within as a work.

lightning run app train_pl.py

lightning run app train_any.py