New error in Trainer started appearing recently in a previously running code.

[cwoolfo1]

When I create a Trainer and run Trainer.fit() I am now getting the following error:

raise TypeError("cannot assign '{}' as child module '{}' "
TypeError: cannot assign 'int' as child module 'precision' (torch.nn.Module or None expected)

This is a new error and this code was just working earlier. Do yall know what could be causing this issue?

[rohitgr7]

hey @cwoolfo1 !

can you share the complete stack trace?

[cwoolfo1]

/.conda/envs/Torch2PyPy/lib/python3.9/site-packages/pytorch_lightning/trainer/connectors/callback_connector.py:58: LightningDeprecationWarning: Setting `Trainer(
stochastic_weight_avg=True)` is deprecated in v1.5 and will be removed in v1.7. Please pass `pytorch_lightning.callbacks.stochastic_weight_avg.StochasticWeightAveraging` directly to the Trai
ner's `callbacks` argument instead.
  rank_zero_deprecation(
GPU available: False, used: False
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
Traceback (most recent call last):
  File "/GRUwithInterpretation/MC4R4.py", line 238, in <module>
    trainer.fit(model, train_dataloader, val_dataloader)
  File "/.conda/envs/Torch2PyPy/lib/python3.9/site-packages/pytorch_lightning/trainer/trainer.py", line 740, in fit
    self._call_and_handle_interrupt(
  File "/.conda/envs/Torch2PyPy/lib/python3.9/site-packages/pytorch_lightning/trainer/trainer.py", line 685, in _call_and_handle_interrupt
    return trainer_fn(*args, **kwargs)
  File "/.conda/envs/Torch2PyPy/lib/python3.9/site-packages/pytorch_lightning/trainer/trainer.py", line 771, in _fit_impl
    self._data_connector.attach_data(
  File "/.conda/envs/Torch2PyPy/lib/python3.9/site-packages/pytorch_lightning/trainer/connectors/data_connector.py", line 190, in attach_data
    self._copy_trainer_model_properties(model)
  File "/.conda/envs/Torch2PyPy/lib/python3.9/site-packages/pytorch_lightning/trainer/connectors/data_connector.py", line 198, in _copy_trainer_model_properties
    m.precision = self.trainer.precision
  File "/.conda/envs/Torch2PyPy/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1212, in __setattr__
    raise TypeError("cannot assign '{}' as child module '{}' "
TypeError: cannot assign 'int' as child module 'precision' (torch.nn.Module or None expected)

[ananthsub]

Do you have an attribute precision defined in your lightning module? If so, this is an improper override of the lightning module which is leading to this error

[cwoolfo1]

That was it! Thank you!

[shreyas301197]

I am not specifying this attribute explicitly in the trainer but still getting error when training on single GPU. What can be the issue? How do you deal with this improper override?

FYI, this error is not happening in multi-GPU training

GPU available: True (cuda), used: True
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
HPU available: False, using: 0 HPUs
Traceback (most recent call last):
File "/home/shreyasve/ai-pers-notebooks/ItemSage/lightning_scripts/main.py", line 220, in
main(model_save_dir, **model_args)
File "/home/shreyasve/ai-pers-notebooks/ItemSage/lightning_scripts/main.py", line 165, in main
trainer.fit( model=model, datamodule=data_module ) # {data,model}_v1.py scripts
File "/opt/zillow/.venv/lib/python3.9/site-packages/pytorch_lightning/trainer/trainer.py", line 608, in fit
call._call_and_handle_interrupt(
File "/opt/zillow/.venv/lib/python3.9/site-packages/pytorch_lightning/trainer/call.py", line 38, in _call_and_handle_interrupt
return trainer_fn(*args, **kwargs)
File "/opt/zillow/.venv/lib/python3.9/site-packages/pytorch_lightning/trainer/trainer.py", line 638, in _fit_impl
self._data_connector.attach_data(
File "/opt/zillow/.venv/lib/python3.9/site-packages/pytorch_lightning/trainer/connectors/data_connector.py", line 148, in attach_data
self._copy_trainer_model_properties(model)
File "/opt/zillow/.venv/lib/python3.9/site-packages/pytorch_lightning/trainer/connectors/data_connector.py", line 153, in _copy_trainer_model_properties
model.precision = int(self.trainer.precision) if self.trainer.precision != "bf16" else "bf16"
File "/opt/zillow/.venv/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1304, in setattr
raise TypeError("cannot assign '{}' as child module '{}' "
TypeError: cannot assign 'int' as child module 'precision' (torch.nn.Module or None expected)

[ananthsub]

Do you have an attribute precision defined in your lightning module? If so, this is an improper override of the lightning module which is leading to this error: https://github.com/PyTorchLightning/pytorch-lightning/blob/49a4a36ad45b937dd0124ecfb08eb7400dbf3950/pytorch_lightning/core/lightning.py#L102-L103