Back to publications
ConferenceAdvances in Neural Information Processing Systems (NeurIPS), 2026

Clapping: Removing Per-sample Storage for Pipeline Parallel Distributed Optimization with Communication Compression

B Kong, X Huang, Y Xu, Y Liang, B Wang, K Yuan

Distributed OptimizationCommunication CompressionMemory-Efficient Training
PDF

Clapping eliminates per-sample storage requirements in pipeline parallel distributed optimization, significantly reducing memory overhead for large-scale training.