method · first seen 1 Nov 2026
chunkwise parallel training
Processing sequences in chunks to expose matrix-level parallelism in recurrent or linear attention models.
Mentions per week
PapersCodeDiscussion
Earliest sources
- ReWY: Faster Backpropagation for Delta-Rule Attention - From 7 to 3 Matrix MultiplicationsPapers · 1 Nov 2026 · Yi Zhang, Zhixin Zhao, Weiwei Sun, Yitao Hu, Sheng Chen, Keqiu Li
Latest sources
- ReWY: Faster Backpropagation for Delta-Rule Attention - From 7 to 3 Matrix MultiplicationsPapers · 1 Nov 2026 · Yi Zhang, Zhixin Zhao, Weiwei Sun, Yitao Hu, Sheng Chen, Keqiu Li