CompAct: Compressed Activations for Memory-Efficient LLM Training
This addresses a major bottleneck in training large language models by targeting the largest memory component, offering incremental improvements over existing methods.
The paper tackles the problem of high GPU memory usage during LLM training by compressing activations, achieving 25-30% memory reduction for pretraining and 50% for fine-tuning.
We introduce CompAct, a technique that reduces peak memory utilization on GPU by 25-30% for pretraining and 50% for fine-tuning of LLMs. Peak device memory is a major limiting factor in training LLMs, with various recent works aiming to reduce model memory. However most works don't target the largest component of allocated memory during training: the model's compute graph, which is stored for the backward pass. By storing low-rank, compressed activations to be used in the backward pass we greatly reduce the required memory, unlike previous methods which only reduce optimizer overheads or the number of trained parameters. Our compression uses random projection matrices, thus avoiding additional memory overheads. Comparisons with previous techniques for either pretraining or fine-tuning show that CompAct substantially improves existing compute-performance tradeoffs. We expect CompAct's savings to scale even higher for larger models.