arXiv:2604.27077v2 Announce Type: replace-cross
Abstract: The Normalized Transformer, or nGPT (arXiv:2410.01131) achieves impressive training speedups and does not require weight decay or learning rate warmup. However, despite having hyperparameters that explicitly scale with model size, we observe that nGPT does not exhibit learning rate transfer across model dimension and token horizon. To rectify this, we combine numerical experiments with a principled use of alignment exponents (arXiv:2407.05872) to revisit and modify the $mu$P approach to hyperparameter transfer (arXiv:2011.14522). The result is a novel nGPT parameterization we call $nu$GPT. Through extensive empirical validation, we find $nu$GPT exhibits learning rate transfer across width, depth, and token horizon.
Disclosure in the era of generative artificial intelligence
Generative artificial intelligence (AI) has rapidly become embedded in academic writing, assisting with tasks ranging from language editing to drafting text and producing evidence. Despite