FastWAM GRPO RL β€” LIBERO-10 (round15 start)

Critic-free flow-SDE GRPO fine-tune of a FastWAM policy (β‰ˆ1B action expert trained; 5B Wan2.2 video expert frozen) on the LIBERO-10 (Long) suite, starting from the round15 SFT checkpoint. Trained with RLinf (FSDP2, 4Γ—A100), grpo_norm_by_std=false (Dr.GRPO), noise_level=0.6, lr 2e-5, clip 1.0, 256 traj/step, 10 steps.

dcp_checkpoint/ is a Torch Distributed Checkpoint (model + optimizer), resumable via RLinf runner.resume_dir.

LIBERO-130 eval (RL step10 vs base round15)

Deterministic full eval β€” all 50 init-states/task, per-suite max-horizon, 100 parallel envs, zero measurement variance. RL was trained only on libero_10; the other suites test generalization / forgetting.

Suite tasks base SR RL SR Ξ”
Spatial 10 0.962 0.972 +0.010
Object 10 0.972 0.912 βˆ’0.060
Goal 10 0.956 0.962 +0.006
Long (libero_10, trained) 10 0.852 0.906 +0.054
Short (libero_90) 90 0.914 0.850 βˆ’0.065
LIBERO-130 (task-weighted) 130 0.921 0.877 βˆ’0.044

GRPO on libero_10 gives a real, headroom-driven gain on the trained suite (+0.054) and holds spatial/goal, but the action expert specializes toward libero_10's long-horizon behavior and mildly forgets object (βˆ’0.060) and the saturated libero_90 (βˆ’0.065; 54/90 tasks worse). Since libero_90 is 90 of the 130 tasks, forgetting dominates the aggregate.

Full per-suite/per-task numbers and reproduction scripts: wzii/RLinf Β· experiments/fastwam_rl_libero10.

Downloads last month

-

Downloads are not tracked for this model. How to track
Video Preview
loading