Instantiating WandBLogger in CLI: assert log_dir is not None

[rbracco]

I've read the docs and checked issues, but I can't figure out how to instantiate WandBLogger using LightningCLI, as the logger save_dir apparently doesn't get passed to the trainer.

results in

new_trainer.py 21 <module>
cli = ExperimentCLI(

cli.py 566 __init__
self._run_subcommand(self.subcommand)

cli.py 837 _run_subcommand
fn(**fn_kwargs)

trainer.py 768 fit
self._call_and_handle_interrupt(

trainer.py 721 _call_and_handle_interrupt
return trainer_fn(*args, **kwargs)

trainer.py 809 _fit_impl
results = self._run(model, ckpt_path=self.ckpt_path)

trainer.py 1172 _run
self._call_setup_hook()  # allow user to setup lightning_module in accelerator environment

trainer.py 1491 _call_setup_hook
self._call_callback_hooks("setup", stage=fn)

trainer.py 1634 _call_callback_hooks
fn(self, self.lightning_module, *args, **kwargs)

cli.py 434 setup
assert log_dir is not None

AssertionError

Because the log_dir isn't set on the trainer. But if I set it in the config as so...

it tells me that it isn't being used anywhere.

Please help.

[kubitz]

Did you solve this issue @rbracco ?

[rbracco]

Unfortunately no, I stopped using the cli for the time being but I plan to go back and try again sometime soon. I'll post here if i find a solution

[tshu-w]

I have made a PR to fix this issue #14878. Currently, you can pass dir to init_args:

init_args:
  save_dir: "logs/"
  dir: "logs/"

[tshu-w]

I have made a PR to fix this issue #14878