Outer-Momentum Restarting in High-Dimensional Two-Phase Optimization
For practitioners of distributed deep learning, this provides a simple mechanism to improve stability and robustness of communication-efficient training.
The paper studies periodic restarting of outer momentum in communication-efficient distributed optimizers like DiLoCo, showing that resets discard stale momentum while preserving inner-loop progress. In language-model pretraining, periodic restarts widen the stable range of outer learning rates and momentum values across communication periods.
Communication-efficient distributed optimizers such as DiLoCo reduce synchronization costs by letting workers perform many local updates before aggregating their progress with an outer momentum optimizer. Recent theory suggests that the outer optimizer acts on an effective spectrum induced by the inner optimization loop, and that the choice of outer momentum controls how progress from local updates is accumulated across communication rounds. We study periodic restarting of the outer momentum as a simple complementary mechanism for controlling this outer memory. In a linearized squared-loss model where prediction-space residuals evolve under the empirical NTK, we derive a mode-wise restart contraction showing that resets exploit phase cancellation by discarding stale momentum while preserving inner-loop progress. Toy experiments verify the predicted contraction behavior, and language-model pretraining experiments show that periodic restarts widen the stable range of outer learning rates and momentum values across communication periods.