DeepSpeed-Ulysses keeps communication volume constant for sequence-parallel attention when sequence length and device count scale together, delivering 2.5x faster training on 4x longer sequences than prior SOTA.
Reducing activation recomputation in large transformer models
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2023 1verdicts
ACCEPT 1representative citing papers
citing papers explorer
-
DeepSpeed Ulysses: System Optimizations for Enabling Training of Extreme Long Sequence Transformer Models
DeepSpeed-Ulysses keeps communication volume constant for sequence-parallel attention when sequence length and device count scale together, delivering 2.5x faster training on 4x longer sequences than prior SOTA.