PyTorch's ezyang: SAC should target norm+attn/mlp, not just the block

ezyang · x · 2026-09-26

PyTorch core developer ezyang discusses Saved Activation Checkpointing (SAC) granularity in transformers.

Related event: PyTorch Maintainer: SAC Can Be Applied at Finer Granularity Than Whole Blocks(2 posts)→

Original post →

More from Research

Research channel →