Core/Thread count doesn't respect cgroup limits

[huston-rogers]

Is there any information or documentation on specifying specific hardware limits or resources to utilize? Trying to use labelme in a batch/HPC environment, and it is flatly failing. It doesn't respect cgroup thread limits, nor does it appear to use the allocated GPUs, even with the onnxruntime-gpu package installed. It hangs nodes on "create ai-polygon". It runs way past the allocated core count, often over the maximum core count.

[huston-rogers]

Bumping discussion. This is a stop work for users who want to utilize labelme on HPC systems. It might be related to items opened under issues recently.