X-UMX: lr_scheduler_step() missing 1 required positional argument: 'metric'

[DavidDiazGuerra]

Hello,

I'm trying to use the receipt for training the X-UMX model on the musdb18 dataset but the following exception raises after finishing the validation of the first epoch:

Traceback (most recent call last):
  File "/opt/tuni/worktmp/repos/asteroid/egs/musdb18/X-UMX/train.py", line 498, in <module>
    main(arg_dic, plain_args)
  File "/opt/tuni/worktmp/repos/asteroid/egs/musdb18/X-UMX/train.py", line 465, in main
    trainer.fit(system)
  File "/worktmp/anaconda3/envs/py39torch20/lib/python3.9/site-packages/pytorch_lightning/trainer/trainer.py", line 520, in fit
    call._call_and_handle_interrupt(
  File "/worktmp/anaconda3/envs/py39torch20/lib/python3.9/site-packages/pytorch_lightning/trainer/call.py", line 42, in _call_and_handle_interrupt
    return trainer.strategy.launcher.launch(trainer_fn, *args, trainer=trainer, **kwargs)
  File "/worktmp/anaconda3/envs/py39torch20/lib/python3.9/site-packages/pytorch_lightning/strategies/launchers/subprocess_script.py", line 92, in launch
    return function(*args, **kwargs)
  File "/worktmp/anaconda3/envs/py39torch20/lib/python3.9/site-packages/pytorch_lightning/trainer/trainer.py", line 559, in _fit_impl
    self._run(model, ckpt_path=ckpt_path)
  File "/worktmp/anaconda3/envs/py39torch20/lib/python3.9/site-packages/pytorch_lightning/trainer/trainer.py", line 935, in _run
    results = self._run_stage()
  File "/worktmp/anaconda3/envs/py39torch20/lib/python3.9/site-packages/pytorch_lightning/trainer/trainer.py", line 978, in _run_stage
    self.fit_loop.run()
  File "/worktmp/anaconda3/envs/py39torch20/lib/python3.9/site-packages/pytorch_lightning/loops/fit_loop.py", line 202, in run
    self.on_advance_end()
  File "/worktmp/anaconda3/envs/py39torch20/lib/python3.9/site-packages/pytorch_lightning/loops/fit_loop.py", line 377, in on_advance_end
    self.epoch_loop.update_lr_schedulers("epoch", update_plateau_schedulers=not self.restarting)
  File "/worktmp/anaconda3/envs/py39torch20/lib/python3.9/site-packages/pytorch_lightning/loops/training_epoch_loop.py", line 300, in update_lr_schedulers
    self._update_learning_rates(interval=interval, update_plateau_schedulers=update_plateau_schedulers)
  File "/worktmp/anaconda3/envs/py39torch20/lib/python3.9/site-packages/pytorch_lightning/loops/training_epoch_loop.py", line 350, in _update_learning_rates
    call._call_lightning_module_hook(
  File "/worktmp/anaconda3/envs/py39torch20/lib/python3.9/site-packages/pytorch_lightning/trainer/call.py", line 142, in _call_lightning_module_hook
    output = fn(*args, **kwargs)
TypeError: lr_scheduler_step() missing 1 required positional argument: 'metric'

I am using Pytorch 2.0.0 and Lightning 2.0.2. In order to make the model work with Pytorch 2.0.0 I've made some minor changes in the X-UMX model (see #662) and I've reduced the batch size to make the model fit in my local GPU, but I don't think I've made any changes that should affect the Lightning training.

Maybe this could be related to be using recent versions of Pytorch and Lightning? I have no experience working with Lightning.

Best regards,
David

[DavidDiazGuerra]

Hi again,

The problem disappeared after downgrading Lightning to 1.9.4. I guess there are some changes in the 2.0 version that breaks the X-UMX training routine of Asteroid.

Best,
David

[LeonieBorne]

Hello 👋

I had the same problem, it seems that the function definition is def lr_scheduler_step(self, scheduler, metric) in version 2.0 instead of def lr_scheduler_step(self, scheduler, optimizer_idx, metric) in previous versions (see the custom learning rate schedulers documentation here).

@mpariente Should we change this and update the requirements to pytorch-lightning>=2.0.0?

[mpariente]

I'm not sure this code has an easy backward compatible fix. So we should

• First change requirements to limit lightning to 1.x versions, and release asteroid 0.6.1,
• Then fix the issue and upgrade to pytorch-lightning>=2.0.0 with asteroid 0.7.0.

[DavidDiazGuerra]

I'm closing this issue since it has been fixed with #682 and the release 0.7.0.