torchrec icon indicating copy to clipboard operation
torchrec copied to clipboard

1.65x qps - remove unnecessary cat

Open SeanXiaohengMao opened this issue 1 year ago • 1 comments

Summary: Research doc: https://docs.google.com/document/d/1nDdQiJDnqJKzjzM3ku__Y5j196uxRVEB00Mj6qAl31k/edit

Run ada model: https://www.internalfb.com/vanguard/serving_test_cases/487129480789691

We can see huge cpu time spend on cat, which is unnecessary for ada cases, we only cat one tensor, should be a no-op. {F1888776749}

Conditionally remove it to improve latency and qps

Reviewed By: 842974287

Differential Revision: D63398565

SeanXiaohengMao avatar Sep 26 '24 18:09 SeanXiaohengMao