“fused_multi_tensor is not installed corrected” && ”Cannot use GPU“

[YingjieQuBio]

Hello, I make sure the environment has the correct pytorch, pytorch-geometric, cuda versions: Python 3.9.16

Name Version Build Channel pytorch-lightning 1.9.3 pypi_0 pypi torch 2.0.0+cu118 pypi_0 pypi torch-cluster 1.6.3+pt20cu118 pypi_0 pypi torch-geometric 2.5.3 pypi_0 pypi torch-scatter 2.1.1+pt20cu118 pypi_0 pypi torch-sparse 0.6.18+pt20cu118 pypi_0 pypi torch-spline-conv 1.2.2+pt20cu118 pypi_0 pypi torchmetrics 1.4.0.post0 pypi_0 pypi

======================================================================

But I am running the Run inference with only receptor instance with the following warning: fused_multi_tensor is not installed corrected fused_rounding is not installed corrected fused_layer_norm is not installed corrected fused_softmax is not installed corrected Loading pretrained weights from D:\Desktop\code\DeepGlycanSite-main\src\unimol_tools\weights\mol_pre_all_h_220816.pt GPU available: True (cuda), used: False TPU available: False, using: 0 TPU cores IPU available: False, using: 0 IPUs HPU available: False, using: 0 HPUs Testing: 0it [00:00, ?it/s] ...... Testing DataLoader 0: 100%|██████████| 1/1 [00:01<00:00, 1.92s/it] ──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────── Test metric DataLoader 0 ──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────── test_loss 0.0 ────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────

======================================================================

I changed 593 lines of code from torch.device('cpu') to torch.device('cuda'), and the problem still exists, even though torch.cuda.is_available() returns True

I would like to know if there is any impact on the normal use of DeepGlycanSite, and how to use GPU acceleration, Thank you!

[Xinheng-He]

Maybe it's a problem with Windows and switching to Linux system can solve it. Of note, using GPU is not too much accelerating because it's just an inference process.

[YingjieQuBio]

Thank you for the suggestion. The Windows system is not suitable, so I switched to the Linux system and the warning of running the Run inference with only receptor was still there: fused_multi_tensor is not installed corrected fused_rounding is not installed corrected fused_layer_norm is not installed corrected fused_softmax is not installed corrected Loading pretrained weights from /raid/quyj/DeepGlycanSite-main/src/unimol_tools/weights/mol_pre_all_h_220816.pt GPU available: True (cuda), used: False ......

================================================================================

but the Retraining DeepGlycanSite could use the GPU without the warning above using Linux system.