Truncate or Not? A Scaling Study of Looped Transformers with Truncated Backpropagation

Xuyuan Xiong, Siyu Chen, Zihan Shao, Yue Su, Weiyang Jin, Ping Luo, Zhuoran Yang, Xihui Liu

arXiv:2610.26030 · 2026-10-29 공개 · arXiv · PDF

scaling-laws model-size looped-transformers validation-loss loop-count truncated-backpropagation training-compute effective-parameters

Abstract

Looped transformers reuse a shared block of layers to increase computational depth without increasing the number of trainable parameters. They are trained with backpropagation through time (BPTT). To save training compute, BPTT is often truncated to the last few loops, but truncation yields biased gradients that may weaken the benefit of looping. In this work, we study the value of truncation in looped transformer training in a systematic way: we jointly vary model size, training tokens, and forward and backward loop counts across nearly 600 training runs and fit a unified scaling law that captures the effects of looping and truncation through effective parameters. Despite the training-compute savings of truncation, our fitted law favors full BPTT when forward and backward loop counts are jointly optimized. In matched-compute runs, full BPTT through fewer loops consistently outperforms truncated settings of the same compute. The law further shows that at fixed model size, the compute-optimal loop count and training-token budget both grow with compute. When model size is also free, the law identifies two fully trained loops as the best looped setting. Experimentally, this setting achieves lower validation loss than the compute-optimal non-looped model with about 27% fewer non-embedding parameters and only 8% more training compute.

한국어 요약

한 줄 요약

Looped Transformer에서 Truncated BPTT의 효과를 규명한 대규모 스케일링 연구.

핵심 기여도

핵심 아이디어

Looped Transformer는 동일 블록을 반복 실행하여 파라미터 수를 유지하면서 계산 깊이를 늘리는 구조다. 그러나 BPTT를 Truncate하면 경사도 편향이 발생해 성능 저하가 우려된다. 본 연구는 600회 이상의 실험을 통해 Forward/Backward Loop Count, 모델 크기, 훈련 토큰 수를 **동시에 조절**하며, **Effective Parameter** 개념을 도입한 통합 스케일링 법칙을 제안한다. 이 법칙은 Truncation이 Loop Gain을 감소시키는 반면, Full BPTT는 Loop Count를 줄이고 Backward Loop를 늘리는 방식으로 더 높은 Effective Parameter를 제공함을 보여준다.

기술적 접근법

주요 결과

의의 및 한계

실용적 활용