PulseAugur
EN
LIVE 13:38:39

New WSqD learning rate schedule offers horizon-free training for large models

By PulseAugur Editorial · [2 sources] ·

Researchers have introduced WSqD, a novel learning rate schedule designed for large model training that is independent of the training horizon. Unlike existing methods like cosine annealing and Warmup-stable-decay (WSD), WSqD's base schedule does not require a predetermined training duration, allowing for flexible extension of training. This approach, inspired by stochastic convex optimization, theoretically achieves optimal convergence rates and has been empirically shown to match or exceed baseline performance on language model pretraining tasks using the SlimPajama corpus. AI

IMPACT This horizon-free learning rate schedule could simplify and improve the efficiency of training large language models.

RANK_REASON The cluster contains an academic paper detailing a new method for large model training.

Read on arXiv stat.ML →

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →

New WSqD learning rate schedule offers horizon-free training for large models

COVERAGE [2]

  1. arXiv stat.ML TIER_1 English(EN) · Jianhao Ma, Yuxin Chen ·

    WSqD: A Horizon-Free Learning Rate Schedule for Large Model Training

    arXiv:2607.10959v1 Announce Type: cross Abstract: Standard learning rate schedules such as cosine annealing are tied to a fixed training horizon, limiting their ability to accommodate post hoc horizon extension. Warmup-stable-decay (WSD) partially addresses this issue by maintain…

  2. arXiv stat.ML TIER_1 English(EN) · Yuxin Chen ·

    WSqD: A Horizon-Free Learning Rate Schedule for Large Model Training

    Standard learning rate schedules such as cosine annealing are tied to a fixed training horizon, limiting their ability to accommodate post hoc horizon extension. Warmup-stable-decay (WSD) partially addresses this issue by maintaining a long constant-rate phase before a short line…

RELATED TOPICS