fix: correct gradient accumulation off-by-one and lr_scheduler over-stepping#82

Merged

anxiangsir merged 3 commits intomainfrom

fix/training-loop-grad-accum-and-lr-scheduler

Feb 10, 2026

Commits on Feb 7, 2026

fix: correct gradient accumulation off-by-one and lr_scheduler over-stepping
Luodian
committed
fix: align scheduler total_iters with optimizer steps under gradient accumulation
Luodian
committed

Commits on Feb 10, 2026

Merge branch 'main' into fix/training-loop-grad-accum-and-lr-scheduler
anxiangsir
authored