WIP: feature(pu): add priorzero init version - #433

Open

puyuan1996 wants to merge 16 commits into

dev-multitask-balance-cleanfrom

dev-multitask-balance-clean-rft

Commits on Oct 18, 2025

tmp
tAnGjIa520
committed
b1efa60

Commits on Oct 20, 2025

Commits on Oct 21, 2025

fix(pu): fix some data type bug
tAnGjIa520
committed
a24cace
fix(pu): fix some type bugs
tAnGjIa520
committed
3758968
feature(pu): add async pipeline support, add off_policy_degree option
tAnGjIa520
committed
028a18e
fix(pu): fix priorzero monitor vars
tAnGjIa520
committed
53803a5
fix(pu): fix mask_padding, obs_shape, polish cot prompts
tAnGjIa520
committed
9d54f08
feature(pu): add priorzero_orz_pipeline
tAnGjIa520
committed
d22f88a
fix(pu): fix vllm engine in priorzero_orz_pipeline
tAnGjIa520
committed
b11d8be

Commits on Oct 22, 2025

fix(pu): fix rft/sft loss, fix raw_obs_text process
tAnGjIa520
committed
0ad3fdf

Commits on Oct 23, 2025

polish(pu): delete unused files
tAnGjIa520
committed
1af10a8
polish(pu): delete unused files
tAnGjIa520
committed
c5fa0fe

Commits on Nov 17, 2025

fix(pu): fix path
wangshulun
committed
4ef57a5