[recipe] feat: asynchronous reward agent with mini-batch pipeline and one-step off-policy training#2854

Open

haolinyan wants to merge 23 commits intoverl-project:mainfrom

Commits on Jul 25, 2025

impl async reward agent
haolinyan
committed

Commits on Jul 28, 2025

add ci test and fix license
haolinyan
committed
fix some format issues
haolinyan
committed

Commits on Jul 30, 2025

fix the support for user-defined reward function and add docs
haolinyan
committed
fix the license header
haolinyan
committed
fix some format issues in docs/
haolinyan
committed
do code format (PEP 8 -> ruff format)
haolinyan
committed
update readme
haolinyan
committed
update img
haolinyan
committed

Commits on Jul 31, 2025

update docs and ray_trainer.py
haolinyan
committed
fix the wording about "OpenAI-style"
haolinyan
committed
update training scripts
haolinyan
committed
rename "update pipeline" as "mini-batch pipeline"
haolinyan
committed
fix ray_trainer.py
haolinyan
committed
update async_reward_agent.svg
haolinyan
committed

Commits on Aug 1, 2025

update authors' information
haolinyan
committed
update check_device_api_usage.py
haolinyan
committed
Update e2e_async_reward_agent.yml
haolinyan
authored
rename logs
haolinyan
committed

Commits on Aug 3, 2025

remove tensorboard artifacts
haolinyan
committed

Commits on Aug 4, 2025

fix: fix the path error in the test script
haolinyan
committed
Merge branch 'volcengine:main' into main
haolinyan
authored
fix: adapt the test script to the latest main branch
haolinyan
committed