0z5a e58a728b80 feat: version rollout weight updates ...

Track a monotonic policy version across optimizer steps, scheduler updates, and rollout results. Serialize synchronous generation with weight acknowledgements and invalidate reusable prefix KV entries so cached samples remain attributable to the behavior policy that generated them.

2026-09-02 19:01:41 +08:00

..

__init__.py

chore : 升级 torch 2.11.0+cu128，移除自定义 Muon，修复 gloo device_id

2026-06-27 16:10:37 +08:00

metric_util.py

refactor: deduplicate low-risk code paths

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

rollout.py

feat: version rollout weight updates

2026-09-02 19:01:41 +08:00

schedule.py

refactor: remove dead code and deduplicate scheduler setup

2026-08-19 14:58:37 +08:00

strategy.py

feat: version rollout weight updates

2026-09-02 19:01:41 +08:00

train_callback.py

fix: publish checkpoints atomically

2026-09-02 15:29:22 +08:00

train_context.py

feat: version rollout weight updates

2026-09-02 19:01:41 +08:00

trainer.py

refactor: rename optimizer step callback hooks to before and after

2026-09-01 14:22:54 +08:00