🆘 Fail to redo the tutorial refit a MEGNet formation energy model with GPU #137
Replies: 1 comment 4 replies
-
|
Seems like your PyTorch was not compiled with CUDA support, as indicated by the error messages? |
Beta Was this translation helpful? Give feedback.
-
|
I can successfully execute the following code with gpu： A = torch.randn(3, 3, requires_grad=True, device="cuda")
B = torch.randn(3, 3, requires_grad=True, device="cuda")
C = torch.matmul(A, B)
C.backward(torch.eye(3, 3, device="cuda"))
print(A.grad)
print(B.grad)If Project Community can provide any of the following or all, that would be great!
|
Beta Was this translation helpful? Give feedback.
-
|
With the same env, it can run well with Pytorch-Fabric-Trainer official demo:
more dig is needed 👀 |
Beta Was this translation helpful? Give feedback.
-
|
Great! Just a note that I don't think a docker image is going to be particularly useful, especially since a lot depends on the specifics of the system (e.g. GPU) you are using. The actual requirements are already spelled out quite clearly and in almost all instances, we have had no trouble just installing from pip in a virtual env (e.g., using conda). |
Beta Was this translation helpful? Give feedback.
-
|
thankfully the main developers have updated the readMe installation instructions, give it a try, I will do as well. Meanwhile the CPU only version is great already |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
env:
I setup a new env with cuda==117(and same output with 118) by conda:
conda create -n matgl python==3.10 conda activate matgl conda install pytorch torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia conda install -c dglteam/label/cu117 dgl pip install matgl ##matgl.0.8.2cuda test is ok by:
Then, redo the tutorial refit a MEGNet formation energy model with small change:
torch.set_default_device("cuda:0")afterimport torchpl.Trainer(max_epochs=20, accelerator="cuda", logger=logger)last run it with python get following output:
according the hinder, add new line to header:
then get same error output 😢
my view:
the code can't pass this line:
trainer.fit(model=lit_module, train_dataloaders=train_loader, val_dataloaders=val_loader)😸 How to get rid of this ?
Thank all of you!
Beta Was this translation helpful? Give feedback.
All reactions