[fsdp, megatron] feat: refactor fully-async and one-step-off training… #5914
e2e_one_step_off_policy.yml
on: push
setup
6s
e2e_one_step_off_policy_fsdp2
3m 20s
e2e_one_step_off_policy_megatron
3m 6s
cleanup
4s