1.5−3.0× lossless training or pre-training speedup. An off-the-shelf, easy-to-implement algorithm for the efficient training of foundation visual backbones.
MIT License
197
stars
9
forks
source link
Perfect job. Can you share cswinBase384_22k model? #1
Thanks for your attention! We have released this ckpt. Please check it at https://github.com/LeapLabTHU/EfficientTrain/blob/main/EVAL.md.