Horace He

Added memory budget to partitioner

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #126370 * __->__ #126320 * #126446 * #126615 * #126613

Add frontend API to memory budget in torch.utils.checkpoint

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #126370 * #126320 * #126446 * #126615 * #126613

Cached required_fw_nodes creation

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #126370 * #126320 * #126446 * __->__ #126613 cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78 @kadeng @chauhang

Prevent partitioner from ever saving views

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #126370 * #126320 * __->__ #126446 * #126615 * #126613 cc @mrshenli @pritamdamania87 @zhaojuanmao @satgera @gqchen @aazzolini @osalpekar @jiayisuse @H-Huang @kwen2501 @awgu @penguinwu...

Fix flexattention not realizing inputs before lowering

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #126370 * #126320 * #126446 * __->__ #126615 * #126613 cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @peterbell10 @ipiszy...

Changed masked out rows logsumexp to be -inf and not zero

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #134650 cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @yf225 @chenyang78 @kadeng @muchulee8 @ColinPeppler @amjames @desertfire @chauhang

Update 2024-08-07-flexattention.md

1

This was a semantic error in the original blog post.