← back
arXivCheng Liang, Tengfei Zhang, Ya Zhang, Yanfeng Wang, Zhouhan LinFri, Oct 30, 2026, 6:29 AM PDT
score 15.2

Looped Transformers Trained With No Extra Compute Cost

Original: UpLoop: Training Looped Transformers via Upcycling at No Extra Pretraining Compute

Source: arxiv.org ↗

Writing ELI5 summary…

Looped Transformers Trained With No Extra Compute Cost · TinyNews · TinyNews