Not compatible with H800 GPU servers

[Ishmael-Yoko]

Running an error on the H800 GPU server

NVIDIA H800 with CUDA capability sm_90 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_37 sm_50 sm_60 sm_70 sm_75 sm_80 sm_86.

The Driver Version is 550.144.03

How to solve it?

Can I upgrade the torch version to 2.4.0?

The version of torch in MolScribe is torch > = 1.11.0, do the related dependencies need to be upgraded after the upgrade?

[alexey-krasnov]

Hi @Ishmael-Yoko ,

I've been using torch versions from 2.4.0 to 2.6.0, and all of them are compatible with Rxnscribe. Also, I use CUDA GPU for Rxnscribe's inference acceleration. You should be aware that your GPU device is seen by torch.
First things firs,t do in your Python environment:

import torch
torch.cuda.is_available()

should be True

Otherwise, please visit:
https://docs.nvidia.com/cuda/cuda-installation-guide-linux/
and follow the installation guide.

Best regards,
Aleksei