0z5a c36846c8a4 fix: condition online DPO on rollout prompts ...

- Concatenate rollout prompts with selected chosen and rejected responses
- Mask prompt tokens from DPO loss while preserving explicit attention visibility
- Cover response selection, padding alignment, and prompt-conditioned inputs

2026-09-02 12:52:50 +08:00

..

config

refactor: simplify training and inference interfaces

2026-08-19 20:55:13 +08:00

dataset

refactor: deduplicate low-risk code paths

2026-08-19 16:17:40 +08:00

extension

refactor: unify operator selection behind generic dispatch

2026-09-01 16:56:57 +08:00

inference

perf: fill steady-state decode input ids via d2d copy

2026-08-31 14:24:51 +08:00

model

feat: load HuggingFace checkpoints via key/config conversion

2026-08-20 11:34:59 +08:00

optim

fix: use torch.optim.AdamW in ManoAdamW instead of NAdamW

2026-08-01 09:20:54 +08:00

parallel

fix: report true rank in logs via dist-aware helpers

2026-08-27 13:39:38 +08:00

preprocessing

fix : correct prefill sampling and record alignment

2026-08-05 22:20:29 +08:00

serialization

refactor: unify CUDA skip guards in tests

2026-08-22 21:08:06 +08:00

tokenize

refactor: simplify BaseFactory and separate ModelFactory from AutoModel

2026-07-30 09:38:20 +08:00

trainer

fix: condition online DPO on rollout prompts

2026-09-02 12:52:50 +08:00

__init__.py

refactor: simplify training and inference interfaces

2026-08-19 20:55:13 +08:00

factory.py

refactor: simplify BaseFactory and separate ModelFactory from AutoModel

2026-07-30 09:38:20 +08:00

logging.py

fix: report true rank in logs via dist-aware helpers

2026-08-27 13:39:38 +08:00

protocols.py

refactor: 重构训练后端为 Executor 模式

2026-05-24 20:35:44 +08:00

signal_handler.py

refactor: move signal_handler from parallel/ to top-level for broader reuse

2026-07-28 10:36:17 +08:00