Anytime Pretraining offers horizon-free LLM training with weight averaging

Researchers have introduced "Anytime Pretraining," a novel approach to training large language models that eliminates the need for pre-defined training horizons. This method utilizes horizon-free learning-rate schedules combined with weight averaging, demonstrating comparable final loss to traditional cosine decay schedules. The findings suggest that this anytime strategy offers a practical and effective alternative for large language model pretraining, particularly in open-ended training scenarios. AI

影响 This research could simplify LLM training by removing the need for complex, horizon-dependent learning rate schedules.

排序理由 The cluster contains a research paper detailing a new method for training large language models. [lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv stat.ML 阅读 →

• Alexandru Meterez
• alphaXiv
• Anytime Pretraining
• arXiv
• CatalyzeX
• Chinchilla
• DagsHub
• Gotit.pub
• Hugging Face
• machine learning
• ScienceCast
• stochastic gradient descent

• 论文
• 基础设施

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →