a5466d4
1
2
3
4
5
6
7
8
Model trained in 8-bit with LoRA Usage: with torch.autocast(): # needed for best performance