vds-4g-utilization-check-20260909-b

molmoact2 policy trained via Vibe Data Studio on 188 teleop episodes.

At a glance

Architecture	molmoact2 (base: allenai/MolmoAct2)
Dataset	cucumber-peel-DAgger-iter2-v2-trim @ 692d1ac8
Scale	188 episodes · 58,994 frames @ 20 fps
Steps / batch / seed	100 / 8 / 42
lerobot	39237fa2
Provider / GPU	gcp / a2-highgpu-4g
Duration	0h 30m
Final train loss	0.070
W&B	vds62-98c60e29

Tasks

Language-conditioned — pass one of these instructions at inference:

Instruction	Episodes
peel the skin of the cucumber with several strokes, starting close to the blue gripper and moving vertically along the cucumber toward the robot origin	188

Episodes per task per recording rig (derived from the nested collection manifests down to leaf recorded datasets):

Task	jetson1	vibepi	Total
peel the skin of the cucumber with several strokes, startin…	117	71	188
Total	117	71	188

Rig

• Action/state (7-dim): shoulder_pan, shoulder_lift, elbow_flex, wrist_flex, wrist_roll, gripper, tilt
• Cameras @ 20 fps: base 640×480, top 640×480, wrist 640×480
• Consumed image inputs (the checkpoint's resolved VISUAL features, from its config.json): observation.images.top, observation.images.wrist, observation.images.base

Data provenance

Materialized collection cucumber-peel-DAgger-iter2@v2 (frozen 2026-07-22):

• cucumber-peel-DAgger-iter0-v1-trim @ 3604420c (42 eps)
• jetson1-peel-doris072026 @ ba099f15 (47 eps)
• vibepi-peel-doris072126 @ 04d8dcf5 (2 eps)
• vibepi-peel-doris072226 @ 8ed1cb6d (42 eps)
• vibepi-peel-adaptive-doris072226 @ 111b736f (27 eps)
• jetson1-peel-adaptive-doris072026 @ a1ad7824 (28 eps)

Reproduce — exact training command

lerobot-train --policy.type=molmoact2 --policy.checkpoint_path=allenai/MolmoAct2 --policy.repo_id=VibeCuisine/vds-4g-utilization-check-20260909-b --dataset.repo_id=VibeCuisine/cucumber-peel-DAgger-iter2-v2-trim --dataset.revision=692d1ac83fea08c85b1b1bdb1bf25e6551f8cf41 --batch_size=8 --steps=100 --save_freq=2000 --output_dir=outputs/train/vds-4g-utilization-check-20260909-b --job_name=vds-4g-utilization-check-20260909-b --policy.device=cuda --wandb.enable=true --wandb.disable_artifact=true --wandb.project=lerobot --wandb.run_id=vds62-98c60e29 --wandb.entity=jeremyhx-freelance --seed=42 --dataset.image_transforms.enable=true --policy.dtype=bfloat16 --policy.action_mode=continuous --policy.inference_action_mode=continuous --policy.control_mode='absolute joint pose' --policy.normalize_gripper=true --policy.setup_type='single robotic arm' --policy.gradient_checkpointing=true --policy.train_mode_vlm=lora --policy.enable_inference_cuda_graph=false --policy.chunk_size=30 --policy.n_action_steps=30 --policy.normalization_mapping='{"ACTION":"QUANTILES","STATE":"QUANTILES","VISUAL":"IDENTITY"}' --policy.scheduler_decay_steps=100 --policy.scheduler_warmup_steps=10 --save_freq=100 --log_freq=10

--num_workers is appended host-side (clamped to the host's allocation).

---

Card generated by Vibe Data Studio on run completion.