0z5a 4019ddac31 perf: reuse rollout behavior logprobs ...

Feed sampler-aligned behavior log-probabilities directly into online GRPO instead of allocating, synchronizing, and forwarding a duplicate old-policy model. Keep the old-model path as an offline compatibility fallback and validate supplied rollout tensors before loss computation.

2026-09-02 19:29:53 +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: accept arbitrary K in bf16 gemv with aligned head-tail sweeps

2026-09-02 14:48:01 +08:00

inference

feat: version rollout weight updates

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

model

perf: accelerate decode linear with bf16 gemv

2026-09-02 13:11:25 +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

fix: publish checkpoints atomically

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

tokenize

refactor: simplify BaseFactory and separate ModelFactory from AutoModel

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

trainer

perf: reuse rollout behavior logprobs

2026-09-02 19:29:53 +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