v0.1.9a2

What's Changed

• [parallel] feat: add enable_forward_prefetch for fsdp2 by @JorgenWan in #660
• [dist] fix: use src_data_rank=None in load_model_weights to skip redundant scatter by @kahlun in #648
• [model] feat: add config.add_helper for module-level helpers in patchgen by @TimYangst in #663
• [model] feat: [transformers v5] support qwen3vl for transformer v5 by @yiwzhao in #527
• [ci] fix: add -v for npu ut e2e test by @phdddd in #667
• [ci] chore: trigger gpu/npu ci only when lint check pass by @FoolPlayer in #650
• [model, ci, agent] feat: migrate qwen3_vl_moe to transformers v5 by @TimYangst in #666
• [model, ci] test: bitwise-equal HF vs veomni logits; fix bf16 MoE dtype divergences by @Luosuu in #670
• [model, ci, agent] feat: migrate qwen3_omni_moe to transformers v5 by @TimYangst in #668
• [task] fix: fix bug for args.train.accelerator.fsdp_config.mixed_precision.enable by @UserChen666 in #638
• [data] chore: add modal mask into keys_to_pad by @JorgenWan in #672
• [docker] feat: add Ascend A3 Dockerfile and docs by @phdddd in #659
• [ops] refactor: replace kernel env vars with args by @FoolPlayer in #651
• [ci] fix: disable pin_memory in test_datasets multisource tests for NPU by @FoolPlayer in #677
• [model] feat: support npu qwen3p5 and qwen3p5-moe by @yicheng-gong in #641
• [model] feat: guard unmigrated models against transformers v5 by @TimYangst in #675
• [misc] feat: support Python 3.12 by @Luosuu in #683
• [docker] feat: Fix npu workflow by @phdddd in #680
• [misc] chore: update fa2 cp312 wheel URL and revert .python-version by @Luosuu in #684
• [docker] fix: Fix workflow npu docker by @phdddd in #686
• [BREAKING][ops, model] feat: registration-based kernel replacement framework by @TimYangst in #678
• [ops, model] refactor: relax moe_implementation Literal to str for extensibility by @TimYangst in #689
• [ci] feat: update ci workflow by @FoolPlayer in #679
• [docs] fix: add install datasets==2.21.0 to ascend doc by @phdddd in #692
• [release] chore: release v0.1.9a2 by @TimYangst in #694

New Contributors

• @kahlun made their first contribution in #648
• @UserChen666 made their first contribution in #638

Full Changelog: v0.1.9a1...v0.1.9a2