Skip to content

v1.2.2

Choose a tag to compare

View all tags
@francois-drielsma francois-drielsma released this 13 Sep 06:21
· 16 commits to main since this release
v1.2.2
d4de053

Added

  • Scheduler restart policy: Allow resumed training to preserve optimizer tensor state while explicitly restarting a newly configured learning-rate schedule; complete scheduler restoration remains the default.

Fixed

  • Iteration-based training limits: Keep positive explicit iteration counts independent from epoch limits so ordinary and resumed runs do not stop after one input pass.

Full Changelog: v1.2.1...v1.2.2

Assets 2
Loading