gradient-accumulation

Tag

Cards List
#gradient-accumulation

Breaking the Bubble: Asynchronous Pipeline Parallel Training with Bounded Weight Inconsistency

Hugging Face Daily Papers · 2026-06-05 Cached

Introduces PACI, a bubble-free asynchronous pipeline parallel training method that bounds forward/backward weight inconsistency using local gradient accumulation, achieving higher throughput and faster time-to-accuracy without sacrificing stability or memory usage.

0 favorites 0 likes
← Back to home

Submit Feedback