Lightning CLI ckpt_path not loading model weights or hyperparams

[mhbakalar]

I am having trouble loading a model from checkpoint using the lightning CLI. I have confirmed that I am calling save_hyperparams() in the model init. I also have no problem loading the model using pure lightning code. However, when calling:
python trainer.py validate --ckpt_path lightning_logs/version_0/checkpoints/epoch=19-step=9940.ckpt, the parameters seem to all have NoneType values. Two snippets below – the successful checkpoint loading using pure lightning, followed by the error message when using lightning CLI. Thanks for the help.

>>> import lightning
>>> from lit_modules import modules
>>> modules.Regression.load_from_checkpoint('lightning_logs/version_0/checkpoints/epoch=19-step=9940.ckpt')
Regression(
  (model): MLPModel(
    (flatten): Flatten(start_dim=1, end_dim=-1)
    (linear_relu_stack): Sequential(
      (0): Linear(in_features=230, out_features=250, bias=True)
      (1): ReLU()
      (2): Dropout(p=0.5, inplace=False)
      (3): Linear(in_features=250, out_features=1, bias=True)
    )
  )
  (loss_fn): MSELoss()
)

(base) matthewbakalar@LM-MBAKALAR cryptic % python trainer.py validate --ckpt_path=lightning_logs/version_0/checkpoints/epoch=19-step=9940.ckpt
/Users/matthewbakalar/anaconda3/lib/python3.11/site-packages/lightning/fabric/utilities/seed.py:40: No seed found, seed set to 4014711880
Seed set to 4014711880
Traceback (most recent call last):
  File "/Users/matthewbakalar/Documents/Code/cryptic/trainer.py", line 55, in <module>
    cli_main()
  File "/Users/matthewbakalar/Documents/Code/cryptic/trainer.py", line 50, in cli_main
    cli = LightningCLI(modules.Regression)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/matthewbakalar/anaconda3/lib/python3.11/site-packages/lightning/pytorch/cli.py", line 383, in __init__
    self.instantiate_classes()
  File "/Users/matthewbakalar/anaconda3/lib/python3.11/site-packages/lightning/pytorch/cli.py", line 533, in instantiate_classes
    self.config_init = self.parser.instantiate_classes(self.config)
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/matthewbakalar/anaconda3/lib/python3.11/site-packages/jsonargparse/_deprecated.py", line 140, in patched_instantiate_classes
    cfg = self._unpatched_instantiate_classes(cfg, **kwargs)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/matthewbakalar/anaconda3/lib/python3.11/site-packages/jsonargparse/_core.py", line 1182, in instantiate_classes
    cfg[subcommand] = subparser.instantiate_classes(cfg[subcommand], instantiate_groups=instantiate_groups)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/matthewbakalar/anaconda3/lib/python3.11/site-packages/jsonargparse/_deprecated.py", line 140, in patched_instantiate_classes
    cfg = self._unpatched_instantiate_classes(cfg, **kwargs)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/matthewbakalar/anaconda3/lib/python3.11/site-packages/jsonargparse/_core.py", line 1176, in instantiate_classes
    component.instantiate_class(component, cfg)
  File "/Users/matthewbakalar/anaconda3/lib/python3.11/site-packages/jsonargparse/_signatures.py", line 577, in group_instantiate_class
    parent[key] = instantiator_fn(group.group_class, **value)
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/matthewbakalar/anaconda3/lib/python3.11/site-packages/jsonargparse/_common.py", line 109, in default_class_instantiator
    return class_type(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/matthewbakalar/Documents/Code/cryptic/lit_modules/modules.py", line 90, in __init__
    self.model = models.MLPModel(
                 ^^^^^^^^^^^^^^^^
  File "/Users/matthewbakalar/Documents/Code/cryptic/models/models.py", line 18, in __init__
    nn.Linear(input_size, hidden_size),
    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/matthewbakalar/anaconda3/lib/python3.11/site-packages/torch/nn/modules/linear.py", line 96, in __init__
    self.weight = Parameter(torch.empty((out_features, in_features), **factory_kwargs))
                            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: empty() received an invalid combination of arguments - got (tuple, dtype=NoneType, device=NoneType), but expected one of:
 * (tuple of ints size, *, tuple of names names, torch.memory_format memory_format, torch.dtype dtype, torch.layout layout, torch.device device, bool pin_memory, bool requires_grad)
 * (tuple of ints size, *, torch.memory_format memory_format, Tensor out, torch.dtype dtype, torch.layout layout, torch.device device, bool pin_memory, bool requires_grad)

[mauvilsa]

I haven't looked at the error details. But it is likely that #18105 is related. That is my proposal to properly support save_hyperparameters and load_from_checkpoint for models trained with LightningCLI. There hasn't been any activity in that pull request for months. You can comment there to give visibility to this topic.