pytorch_lightning.utilities.exceptions.MisconfigurationException
vincenttsai2015 opened this issue · comments
As I was executing the command "python3 guidance/train_qm9_regressor.py +experiment=regressor_model.yaml", an error message "pytorch_lightning.utilities.exceptions.MisconfigurationException: The provided lr scheduler ExponentialLR doesn't follow PyTorch's LRScheduler API. You should override the LightningModule.lr_scheduler_step hook with your own logic if you are using a custom LR scheduler." occurred.
I'm wondering if I installed the wrong version of Pytorch (2.0 with CUDA 11.8) and Pytorch-geometric.
Is it possible to solve this by downgrading the version of Pytorch or some other module? Thanks.
[Update 1]
As I tried to train a regressor after installing the guidance version according to the instruction (with Pytorch 1.11+CUDA11.3) of readme, a runtime error occurred: RuntimeError: object has no attribute sparse_csc_tensor
[Update 2]
With Pytorch 1.10+CUDA 11.1, another error occurred as I tried to train a regressor after installing the guidance version according to the instruction: AttributeError: module 'distutils' has no attribute 'version'
I found the common solution is downgrading the version of setuptools, but if I try to downgrade the version of setuptools to <=59.5.0, running the code "train_qm9_regressor.py" will lead to "Segmentation fault (core dumped)"......
Is there anything else I can try? Thanks.
Hello, the code was developed with torch_geometric==1.7 and cuda=11.3.
You can try to downgrade to these versions. Before doing that, check that the segmentation fault is not related to the same problem as #48.
I have tried the latest version of Pytorch, Pytorch Geometric, and Pytorch lightning equipped with CUDA 11.8, and solved this error after considerable modifications. Thanks.