main
Acc fix in xpu (#47500)
Update distributed tests (#44338)
[Test refactor 1/5] Per-folder tests reorganization (#15725)
Allow position_ids_start=2 on DataCollatorWithFlattening for RoBERTa etc. (#47525)
Fix loss alignment and Trainer token counting for encoder decoder models (#46903)
Update common tests Trainer (#44260)
Fix `TrackioCallback` fails to log evaluation metrics after training ends (#46935)
Use explicit nn.Buffer everywhere for modular (#47722)
Improve Trainer DataLoader Controls for Streaming and Multiprocessing (#47164)
fix-trainer-tests (#46541)
Add GreedyLR adaptive learning rate scheduler (#44271)
Fix training ci and clean some tests (#44491)
feat(integration): Add KubeflowCallback to enable automatic progress … (#44487)
Ensure final evaluation runs with step-based evaluation strategy (#44146)