Claire Radar

OmniOpt

原始来源英文简介

A cross-domain benchmark for comparing optimizers in large-scale model training. It covers 24+ optimizers across two stages: Stage 1 sweeps on C4 with LLaMA-3 architectures (60M to 1B), and Stage 2 transfers to FineWeb-Edu with four architectures (Transformer++, GLA, DeltaNet, Gated DeltaNet) at 340M and 1B scales. Controlled-variable protocol with fixed architecture, data, and schedule settings.