datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
gsworld_pour_mustard_sft_100
GSWorld FR3 PourMustard SFT 100
LeRobot v2.1 dataset converted from GSWorld / ManiSkill motion-planning demonstrations.
Task: pour the mustard bottle onto the bread slice
Robot: fr3_umi
Episodes: 100
Frames: 22041
FPS: 20
Schema: GSWorld dual-camera, DROID-style OpenPI fields
Cameras: exterior_image_1_left from GSWorld right_cam, wrist_image_left from GSWorld wrist_cam
State: 7 arm joint positions + 1 gripper position
Action: 8D FR3 UMI env-native pd_joint_pos
OpenPI reproduction… See the full description on the dataset page: https://huggingface.co/datasets/gst0312/gsworld_pour_mustard_sft_100.gsw-eval
gsw-eval — Swiss-German evaluation benchmark
A small, honest benchmark over held-out text from
mischeiwiller/swiss-german-text.
Built by scripts/build_eval_split.py (task 6.1) via the pure mundart.eval layer.
Why not regional dialect-ID? Every row of the source corpus carries
region="unknown" — no greenlit source supplies a defensible dialect-region
label — so a regional classification task would require inventing labels we
cannot defend. The task the data genuinely supports is… See the full description on the dataset page: https://huggingface.co/datasets/mischeiwiller/gsw-eval.gsworld_stack_sft_100
GSWorld FR3 Stack SFT 100
LeRobot v2.1 dataset converted from GSWorld / ManiSkill motion-planning demonstrations.
Task: stack the red tomato can on top of the tomato soup can
Robot: fr3_umi
Episodes: 100
Frames: 16699
FPS: 20
Schema: GSWorld dual-camera, DROID-style OpenPI fields
Cameras: exterior_image_1_left from GSWorld right_cam, wrist_image_left from GSWorld wrist_cam
State: 7 arm joint positions + 1 gripper position
Action: 8D FR3 UMI env-native pd_joint_pos
OpenPI… See the full description on the dataset page: https://huggingface.co/datasets/gst0312/gsworld_stack_sft_100.