Willmish
commented
5 months ago - [ ] Script tuned for scheduling unlearing of specified task, according to the plan: https://github.com/Adamliu1/SNLP_GCW/issues/98 (128, 512, 1024 sequential for splits 4, 16, 64; 128, 512, 1024 batch; 20480 samples continuous)
- [ ] Tuned to be ran on a 4 GPU node (could be expanded/adapted to fewer, but requires manual tweaking)