0% GPU utilization; torch.cuda.is_available() reports true #3579

Is there an existing issue for this?

• I have searched the existing issues

Operating System

Windows 11

DeepLabCut version

3.0.0

What engine are you using?

pytorch

DeepLabCut mode

single animal

Device type

Nvidia GeForce RTX 5070 Laptop GPU

Bug description 🐛

We were wondering why our training was taking so long (sometimes up to a week!) when it finally occurred to us to check task manager. Lo and behold, the training was running entirely on the CPU with 0% GPU utilization. After some poking around, I discovered that DLC was not at first installed properly. Anaconda had been installed, but PyTorch and CUDA had not. I fixed those issues, I got torch.cuda.is_available() to report true, and for good measure I deleted the Deeplabcut environment and reinstalled it according to the instructions on the install page. No luck--task manager continued to report a totally inactive GPU when I started a test training run while the CPU was chugging away.

Steps To Reproduce

nvidia-smi output:
SMI.txt
Project YAML:
config.yaml
No. of Videos: 12

Network Architecture: Have tried resnet_50 and resnet_101

PyTorch Version: 2.14.1 (from conda-forge)

Git Version: 2.56.0.4

That's everything I can think to add. Tell me if you need anything else!

Relevant log output

Anything else?

No response

Code of Conduct

• I agree to follow this project's Code of Conduct