How to log pruning sparsity in wandb logger

[whtitefall]

Hi,

I'm trying to use model pruning in lightning, I know the verbose parameter log out the sparsity of the model layers,

but how do I log the sparsity versus epoch/steps into the logger? is there a customized logger function that I need to specify?

wandb_logger = pl.loggers.wandb.WandbLogger()

trainer = pl.Trainer.from_argparse_args(args, 
    logger=wandb_logger,
    callbacks=[ModelPruning("l1_unstructured", amount=0.2,verbose=2,)]
    )
trainer.fit(model, dm)

[rohitgr7]

hey!
logging pruning stats isn't supported natively since it's an extension,, but you can use self.logger.experiment.log to log any kind of stats supported by wandb. To access the step you can use self.global_step/self.current_epoch.