Output Embedding Centering for Stable LLM Pretraining

arXiv:2601.02031v3 Announce Type: replace-cross Abstract: Pretraining of large language models is not only expensive but also prone to certain training instabilities. A specific instability that often occurs at the end of training is output logit divergence. The most widely used mitigation strategies, z-loss and logit soft-capping, merely address the symptoms rather than the underlying cause of the problem. In this paper, we analyze the instability from the perspective of the output embeddings' geometry and identify anisotropic embeddings as its source. Based on this, we propose output embedding centering (OEC) as a new mitigation strategy, and demonstrate that it suppresses output logit divergence. OEC can be implemented in two different ways: as a deterministic operation called mu-centering, or a regularization method called mu-loss. Our experiments show that both variants outperform z-loss in terms of training stability, while being on par with logit soft-capping. This holds true both in the presence and the absence of weight tying. As a secondary result, we find that mu-loss is significantly less sensitive to regularization hyperparameter tuning than z-loss.

Related

• Efficient Pre-Training with Token Superposition
• ReGATE: Learning Faster and Better with Fewer Tokens in MLLMs
• Is More Data Worth the Cost? Dataset Scaling Laws in a Tiny Attention-Only Decoder

Source: arXiv cs.CL | 2026-09-11