variable task.gpus like task.cpus.

[alessiovignoli]

New feature

Helllo. I could not find in the documentation a variable that hold the number of GPUs used by the process,. The closest i could find is this -> task.accelerator.request from here.

It would be nice to have a variable called task.gpus like there is for the cpu task.cpus.

Usage scenario

At running time of the process it can be useful to know the number of CPUs,GPUs and memory requested. The first and last have their own task variable associated with an intuitive name.

This could be useful metrics to know for GPU intensive processes like ML applications.
A practical application could be in the case of ray.init() documentation, where an explicit hard limit can be set for resources.

[bentsherman]

Does task.accelerator.request not work for you?