Hi, we use cell2location on a dataset generated by our own Sptial protocol however we have far much more SPOTs than 10X visum , so our GPU ran out of memory, is there any half precision(FP16) api since you built your package based on pytorch, it would be very helpful if you told me how to train this model at half precision level
See the issue about multi GPU training about potential solutions to this problem. If you try half precision - please let us know if it works. It didn’t work for us in the past due to numerical accuracy.
Hi, we use cell2location on a dataset generated by our own Sptial protocol however we have far much more SPOTs than 10X visum , so our GPU ran out of memory, is there any half precision(FP16) api since you built your package based on pytorch, it would be very helpful if you told me how to train this model at half precision level