04:00
2026-09-23
machinebrief.com
large-language-models
Terminal Shrinkage Averaging Reveals a Schedule-Estimator Interaction in LLM Pretraining
A new arXiv paper (2609.25482v1) proposes Terminal Shrinkage Averaging (TSA), a pretraining estimator that interpolates between the raw final iterate and the average of recent checkpoints to balance r…