Is it possible to know how many GPUs are available on the ray cluster on the driver side, prior to allocating work load?

[ziwen_zhang]

I want to know how many GPUs are available on my ray cluster, so I can allocate jobs dynamically.