ketos test has a --workers option but parallelization only happens at the data loader level. I guess it's because the tests used to be performed on GPU but it's not the case anymore. I was wondering if there was a good reason now for not using a process pool.
The middle-term plan is to factor out the testing into the proper pytorch-lightning routines which would make process-level parallelization somewhat difficult. That basically is the main reason.
ketos test
has a--workers
option but parallelization only happens at the data loader level. I guess it's because the tests used to be performed on GPU but it's not the case anymore. I was wondering if there was a good reason now for not using a process pool.