Files
Paul Adenot 5ee025414f Bug 2069895 - Release the speech recognition model's host weights once uploaded. r=media-playback-reviewers,alwu
When Metal is used, the tensors are uploaded to the device in use. We don't
Need to keep the host copy.

This does wonder to RSS (roughly divides by two).

Differential Revision: https://phabricator.services.mozilla.com/D324041
2026-09-10 16:17:14 +00:00
..