AUTOMATIC 254da5d127 --opt-split-attention now on by default for torch.cuda, off for others (cpu and MPS; because the option does not work there according to reports)

2022-09-21 09:49:02 +03:00

15 KiB

Raw Blame History

View Raw