Open iqddd opened 4 months ago
BOFT-training switches to float32? Judging by the speed of training (compared to Diag-OFT), the speed of sample generation and the size of the final file - yes. Is it possible to make it work in mixed precision?
Try to ask the question on the LyCORIS repo. The author will be the best to provide an answer.
BOFT-training switches to float32? Judging by the speed of training (compared to Diag-OFT), the speed of sample generation and the size of the final file - yes. Is it possible to make it work in mixed precision?