Remove force_gpu from training parameters

We should remove the force_cpu parameter, and instead look at available devices. See https://gitlab.com/teklia/atr/dan/-/blob/66f0f92db68d23c1312c617c8556169b5423e18c/dan/manager/training.py#L130

changed milestone to %DAN-P2: Clean code and remove unused code

added P2 label

assigned to @melodie.boillet

After discussion with @starride, we want to keep this parameter because it is useful when we want to debug or run tests, so I close this issue.

closed

unassigned @melodie.boillet