fix: timing_raw

feat: return_dict

refactor: remove redundant response_mask computation (#792)

feat: loss_agg_mode

doc: upgrade to vllm 0.8.2 (#769)

chore: improve news

fix: top p for full version

fix: validation top p as paper

fix: remove uncessary verify from naive

fix: config

fix: algo job name

fix: config

[recipe] refactor: decouple DAPO (#790)

chore: format

feat: CI for DAPO

feat: decouple entrypoint

fix ut

In the GRPO example scripts, modify the entropy_coeff to 0 to ensur…

prevent division by zero in SFT (#786)

fix: prevent NaN when all items are intentionally masked by adding ep…

chore: wandb run of an early version

fix: set torch dtype to auto (#749)

fix: prompt_token_ids should be list[int] instead of np.array (#772)

e2e

Update math_dataset.py to fix typo in the annotation (#765)

fix ci

upd test script