paper-with-me

홈 › Papers

Don't be lazy: CompleteP enables compute-efficient deep transformers

2025-05-02 · Nolan Dey, Bin Claire Zhang, Lorenzo Noci, Mufan Li, Blake Bordelon, Shane Bergsma, Cengiz Pehlevan, Boris Hanin, Joel Hestness

We study compute efficiency of LLM training when using different parameterizations, i.e., rules for adjusting model and optimizer hyperparameters (HPs) as model size changes. Some parameterizations fail to transfer optimal base HPs (such as learning rate) across changes in model depth, requiring practitioners to either re-tune these HPs as they scale up (expensive), or accept sub-optimal training when re-tuning is prohibitive. Even when they achieve HP transfer, we develop theory to show parameterizations may still exist in the lazy learning regime where layers learn only features close to their linearization, preventing effective use of depth and nonlinearity. Finally, we identify and adopt the parameterization we call CompleteP that achieves both depth-wise HP transfer and non-lazy learning in all layers. CompleteP enables a wider range of model width/depth ratios to remain compute-efficient, unlocking shapes better suited for different hardware settings and operational contexts. Moreover, CompleteP enables 12-34% compute efficiency improvements over the prior state-of-the-art.

📄 PDF Abstract BibTeX arXiv:2505.01618

Code (1)

eleutherai/nanogpt-mup 공식 구현 pytorch

Methods 이 논문이 사용한 방법론

BASE 설명 없음
ADOPT Please enter a description about the method here

Similar Papers 제목 키워드 기반

LazyDiT: Lazy Learning for the Acceleration of Diffusion Transformers

2024-12-17 · Xuan Shen, Zhao Song, Yufa Zhou, Bo Chen 외

Diffusion Transformers have emerged as the preeminent models for a wide array of generative tasks, demonstrating superior performance and efficacy across various applications. The promising results come at the cost of sl…

Denoising

LazyFormer: Self Attention with Lazy Update

2021-02-25 · Chengxuan Ying, Guolin Ke, Di He, Tie-Yan Liu

Improving the efficiency of Transformer-based language pre-training is an important task in NLP, especially for the self-attention module, which is computationally expensive. In this paper, we propose a simple but effect…

LazyDrag: Enabling Stable Drag-Based Editing on Multi-Modal Diffusion Transformers via Explicit Correspondence

2025-09-15 · Zixin Yin, Xili Dai, Duomin Wang, Xianfang Zeng 외 arxiv

The reliance on implicit point matching via attention has become a core bottleneck in drag-based editing, resulting in a fundamental compromise on weakened inversion strength and costly test-time optimization (TTO). This…

Image Editing

How Transformers Get Rich: Approximation and Dynamics Analysis

2024-10-15 · Mingze Wang, Ruoxi Yu, Weinan E, Lei Wu

Transformers have demonstrated exceptional in-context learning capabilities, yet the theoretical understanding of the underlying mechanisms remains limited. A recent work (Elhage et al., 2021) identified a ``rich'' in-co…

In-Context Learning

Explaining Grokking in Transformers through the Lens of Inductive Bias

2026-02-06 · Jaisidh Singh, Diganta Misra, Antonio Orvieto arxiv

We investigate grokking in transformers through the lens of inductive bias: dispositions arising from architecture or optimization that let the network prefer one solution over another. We first show that architectural c…