|
I just realized a few lines of code and GPU arrays can be fully async too, with the transfers to CPU only at the end for all of them :D cupy, torch and jax all implement __cuda_array_interface__ Check if __cuda_array_interface__ is implemented, and if so treat that similar to a future. To resolve it just convert to cupy and send to CPU! |
Sorry, something went wrong.
First time I'm doing async stuff, this is an adventure.