Skip to content
中文EN

Data & Training

Training methodologies: dataset curation, synthetic data generation, pre-training and post-training architectures, and compute efficiency.

162 featured itemsRelated TopicsResearch PapersInfra & EngineeringModel Releases

Latest Featured

Items 161–162 of 162
6月27日周一

  1. Softmax Linear Units

    Editorial Pick: 该研究提出了一种新的激活函数SoLU，旨在提高Transformer模型的可解释性，通过实验验证了其在保持性能的同时提高了模型中可解释神经元的比例。

    Transformer CircuitsResearchResearchAgentFeaturedScore 70
7月11日周四