Open t4rf9 opened 2 weeks ago
We're seeing this too, this broke all of TorchGeo's tests: https://github.com/microsoft/torchgeo/actions/runs/9522133755/job/26251028463?pr=2119
+1
We are seeing this as well https://github.com/IBM/terratorch/issues/26
As far as I can tell it stems from https://github.com/Lightning-AI/pytorch-lightning/pull/19771 which (inadvertedly?) affects the LightningCLI parser
Still broken in 2.3.1, still preventing TorchGeo from supporting newer versions of Lightning.
Still broken in 2.3.2.
Bug description
With yaml config file for LightningCLI,
self.save_hyperparameters()
in__init__
of the model and datamodule mistakenly saves adict
containing keys likeclass_path
andinit_args
.This problems appears in version 2.3.0, but version 2.2.5 works correctly.
What version are you seeing the problem on?
2.3.0
How to reproduce the bug
config.yaml
model.py
datamodule.py
main.py
Run
python main.py fit --config config.yaml
- Lightning Component (e.g. Trainer, LightningModule, LightningApp, LightningWork, LightningFlow):
- PyTorch Lightning Version (e.g., 1.5.0):
- Lightning App Version (e.g., 0.5.2):
- PyTorch Version (e.g., 2.0):
- Python version (e.g., 3.9):
- OS (e.g., Linux):
- CUDA/cuDNN version:
- GPU models and configuration:
- How you installed Lightning(
conda
,pip
, source):- Running environment of LightningApp (e.g. local, cloud):