Running validation sanity check only

Trainer
1

Hi all, I’m new to pytorch lightning here.

I want to get all of the logged value before actually running the training, as I understand it currently one way to do this is to run validation sanity check. But I could not find any documentation stating how to run it without Trainer.fit().
is it possible to run only the validation sanity check without running Trainer.fit()? or is there a better way to get all the logged value from LightningModule without running the entire training pipeline?

2

you need something like this> https://github.com/PyTorchLightning/pytorch-lightning/pull/4945

3

Hi, thanks for your answer.
I have quickly scanned the PR and I assumed the feature was still on the development phase and will be released on the v1.2.
if I was going to use the Traner.validate feature, how can I limit the total number of steps done to a small value, like in sanity checks with 2steps?

4

for that, you need this https://pytorch-lightning.readthedocs.io/en/latest/trainer.html#limit-val-batches