paper-with-me

Papers

Spike No More: Stabilizing the Pre-training of Large Language Models

2023-12-28 · Sho Takase, Shun Kiyono, Sosuke Kobayashi, Jun Suzuki

Loss spikes often occur during pre-training of large language models. The spikes degrade the performance of large language models and sometimes ruin the pre-training. Since the pre-training needs a vast computational budget, we should avoid such spikes. Based on the assumption that the loss spike is caused by the sudden growth of the gradient norm, we explore factors to keep the gradient norm small through an analysis of the spectral norms of the Jacobian matrices for the sub-layers. Our findings suggest that stabilizing the pre-training process requires two conditions: small sub-layers and large shortcut. We conduct various experiments to empirically verify our theoretical analyses. Experimental results demonstrate that methods satisfying the conditions effectively prevent loss spikes during pre-training.

📄 PDF Abstract BibTeX arXiv:2312.16903

Code (0)

등록된 구현이 없습니다.

Tasks

Language ModellingLarge Language Model

Methods 이 논문이 사용한 방법론

Focus 설명 없음

Similar Papers 제목 키워드 기반

Stable and low-precision training for large-scale vision-language models

2023-04-25 · NeurIPS 2023 11 · Mitchell Wortsman, Tim Dettmers, Luke Zettlemoyer, Ari Morcos 외

We introduce new methods for 1) accelerating and 2) stabilizing training for large language-vision models. 1) For acceleration, we introduce SwitchBack, a linear layer for int8 quantized training which provides a speed-u…

GPU

Enhancing LLM Training via Spectral Clipping

2026-03-15 · Xiaowen Jiang, Andrei Semenov, Sebastian U. Stich arxiv

While spectral-based optimizers like Muon operate directly on the spectrum of updates, standard adaptive methods such as AdamW do not account for the spectral structure of weights and gradients, leaving them vulnerable t…

SoftSignSGD(S3): An Enhanced Optimizer for Practical DNN Training and Loss Spikes Minimization Beyond Adam

2025-07-09 · Hanyang Peng, Shuang Qin, Yue Yu, Fangqing Jiang 외 arxiv

Adam has proven remarkable successful in training deep neural networks, but the mechanisms underlying its empirical successes and limitations remain underexplored. In this study, we demonstrate that the effectiveness of …

Stochastic Optimization

Stabilizing Reinforcement Learning for Diffusion Language Models

2026-03-06 · Jianyuan Zhong, Kaibo Wang, Ding Ding, Zijin Feng 외 arxiv

Group Relative Policy Optimization (GRPO) is highly effective for post-training autoregressive (AR) language models, yet its direct application to diffusion large language models (dLLMs) often triggers reward collapse. W…

Reinforcement Learning

Stabilizing synchrony by inhomogeneity

2015-03-20

We show that for two identical neuronal oscillators with strictly positive phase resetting curve, isochronous synchrony is an unstable attractor and arbitrarily weak noise can destroy entrainment and generate intermitten…