Does LightningCLI support user-defined subcommands?

[cgoliver]

I would like to define some new subcommands along with the default ones and be able to attach some function calls to it. The documentation does not seem to show this use case so I'm wondering if this is possible.

For example, to visualize the output of some model.

python main.py visualize --checkpoint=/my/checkpoing.ckpt

Or would it be best to just customize the behaviour of some existing subcommands as described here?

[mauvilsa]

To define new subcommands I would recommend the following. First you subclass Trainer and add methods for each of the subcommands you want. These new methods should be usable independent of the CLI. Then subclass LightningCLI, in the __init__ make your new trainer the default and to add the subcommands simply override/extend https://github.com/Lightning-AI/lightning/blob/7a617ec90e1566c763be8ac7a200af1e4025412c/src/pytorch_lightning/cli.py#L425-L434

[mauvilsa]

The user wants to have the output of the model (a batch_size by dim embeddings) clustered in order to use them.

I did not understand this.

[cgoliver]

That was just intended as an example of a subcommand that would do more than just output the raw model predictions. It was confusing and have removed it from the question. Thanks for the answer!