Partial Parameter Updates for Efficient Distributed Training
This addresses communication bottlenecks in distributed training for large-scale models, offering incremental improvements over existing low-communication approaches.
The paper tackles the problem of high communication costs in distributed training by introducing a method that restricts backpropagation to update only a subset of parameters locally, reducing memory usage and FLOPs. Experiments on a 1.3B-parameter language model across 32 nodes show it matches perplexity of prior methods while cutting training FLOPs and peak memory.
We introduce a memory- and compute-efficient method for low-communication distributed training. Existing methods reduce communication by performing multiple local updates between infrequent global synchronizations. We demonstrate that their efficiency can be significantly improved by restricting backpropagation: instead of updating all the parameters, each node updates only a fixed subset while keeping the remainder frozen during local steps. This constraint substantially reduces peak memory usage and training FLOPs, while a full forward pass over all parameters eliminates the need for cross-node activation exchange. Experiments on a $1.3$B-parameter language model trained across $32$ nodes show that our method matches the perplexity of prior low-communication approaches under identical token and bandwidth budgets while reducing training FLOPs and peak memory.