MicroAdam: Accurate Adaptive Optimization with Low Space Overhead and Provable Convergence
This addresses memory constraints for training large-scale models like BERT and LLaMA, offering an incremental improvement over existing optimizers.
The paper tackles the problem of high memory overhead in adaptive optimizers like Adam by proposing MicroAdam, which compresses gradient information and uses compressed error feedback to reduce memory usage while maintaining theoretical convergence guarantees. The result is practical convergence competitive with uncompressed Adam on models like BERT and LLaMA, with lower memory usage and similar running time.
We propose a new variant of the Adam optimizer called MicroAdam that specifically minimizes memory overheads, while maintaining theoretical convergence guarantees. We achieve this by compressing the gradient information before it is fed into the optimizer state, thereby reducing its memory footprint significantly. We control the resulting compression error via a novel instance of the classical \emph{error feedback} mechanism from distributed optimization in which *the error correction information is itself compressed* to allow for practical memory gains. We prove that the resulting approach maintains theoretical convergence guarantees competitive to those of AMSGrad, while providing good practical performance. Specifically, we show that MicroAdam can be implemented efficiently on GPUs: on both million-scale (BERT) and billion-scale (LLaMA) models, MicroAdam provides practical convergence competitive to that of the uncompressed Adam baseline, with lower memory usage and similar running time. Our code is available at https://github.com/IST-DASLab/MicroAdam.