fail500_ps)failure_progress_penalty_scale=500).| Stage | Content |
|---|---|
| returns | meta/*/returns_groot_aiha_multitask_fail500_ps.parquet |
| value SFT | value/global_step_18000/... |
| advantages | meta/*/advantages_groot_aiha_multitask_fail500_ps_N16_ckpt18000_q30.parquet |
| CFG SFT (QAT) | cfg/global_step_*/... |
sm-park/aiha_multitask_groot (GR00T N1.5).export GROOT_RW_VALUE_CKPT="/path/to/this-repo/value/global_step_18000/actor/model_state_dict"1export GROOT_RW_MODEL_PATH="/path/to/this-repo/cfg/global_step_<STEP>/actor"
2# RLinf resolves actor/model_state_dict/full_weights.pt under model_path.meta/*.parquet into:sm-park/multitask_combined_v3 (SFT)sm-park/groot_multitask_FP_rolloutsm-park/groot_multitask_PTQ_rolloutmanifest.csv for exact paths and sizes.