Create an accelerator
accelerator(
device_placement = TRUE,
cpu = FALSE,
cuda_index = torch::cuda_current_device()
)
(logical) whether the accelerator
object should
handle device placement. Default: TRUE
(logical) whether the training procedure should run on the CPU.
(integer) index of the CUDA device to use if multiple GPUs are available. Default: the result of torch::cuda_current_device().