Repository navigation

fix: preserve training state across checkpoints and shared adapters - #315

Merged

Yunnglin merged 4 commits into

mainfrom

fix/training-state-restoration

Oct 15, 2026

Merged

fix: preserve training state across checkpoints and shared adapters#315
Yunnglin merged 4 commits into
mainfrom
fix/training-state-restoration

Commits on Oct 14, 2026
- fix: preserve training state across checkpoints and shared adapters
  Yunnglin
  committed
  19d1849
- fix: use supported Transformers config imports in Megatron strategy
  Yunnglin
  committed
  546d8d1
Commits on Oct 15, 2026
- fix: select one multimodal template hook per forward
  Yunnglin
  committed
  4dc1346
- fix: drop unused sequence parallel optimizer interface expansion
  Yunnglin
  committed
  9e855f0