The Surprising Agreement Between Convex Optimization Theory and Learning-Rate Scheduling for Large Model Training
This provides a theoretical justification for practical learning-rate scheduling in large-scale AI training, though it is incremental as it builds on existing optimization theory.
The paper demonstrates that learning-rate schedules for large model training align closely with a performance bound from non-smooth convex optimization theory, and exploits this match to improve training of 124M and 210M Llama-type models by extending schedules and transferring optimal learning-rates.
We show that learning-rate schedules for large model training behave surprisingly similar to a performance bound from non-smooth convex optimization theory. We provide a bound for the constant schedule with linear cooldown; in particular, the practical benefit of cooldown is reflected in the bound due to the absence of logarithmic terms. Further, we show that this surprisingly close match between optimization theory and practice can be exploited for learning-rate tuning: we achieve noticeable improvements for training 124M and 210M Llama-type models by (i) extending the schedule for continued training with optimal learning-rate, and (ii) transferring the optimal learning-rate across schedules.