Downloads · 30 days
6
9% of all-time downloads
gr00t-n1_5-sponge-marker-delta-25k
mkpongm/gr00t-n1_5-sponge-marker-delta-25k
gr00t-n1_5-sponge-marker-delta-25k is a machine learning model from mkpongm. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
Joints are DELTA; gripper is ABSOLUTE. Action norm: meanstd (joints) + minmax (gripper). Recipe: freeze-LLM (tunellm=False, visual/projector/diffusion tuned), batch 32, 30k steps, decord video backend.
All-time downloads
64
Public
Parameters
2.7B
7.6 GB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors7.6 GB · 100%
Parameter types
How the weights are stored.
BF161.7B · 61%
Model card
From the Hugging Face model README
gr00t N1.5 sponge-marker - DELTA actions (step 25000)
Joints are DELTA; gripper is ABSOLUTE. Action norm: mean_std (joints) + min_max (gripper).
Recipe: freeze-LLM (tune_llm=False, visual/projector/diffusion tuned), batch 32, 30k steps,
decord video backend.
Serving (REQUIRED)
This checkpoint cannot be served with the absolute config - doing so emits raw deltas as
absolute joint targets (measured error up to 65 degrees). Copy so101_delta_configs.py
to the gr00t repo root (that is the only file needed), then:
python scripts/inference_service.py --server \
--model-path mkpongm/gr00t-n1_5-sponge-marker-delta-25k \
--data-config so101_delta_configs:So101SpongeMarkerDeltaDataConfig \
--embodiment-tag new_embodiment --denoising-steps 8 --port 5556
Normalization stats are read from this checkpoint's experiment_cfg/metadata.json like any
other gr00t config - no side-car stats file is needed. action.single_arm holds delta
stats (mean ~= [0.12,-1.63,-2.36,-0.54,0.02]); action.gripper and state.* hold absolute
stats. The client (eval_lerobot.py) needs no changes - unapply restores absolute joint
targets.
- Type
- gr00t_n1_5
- License
- apache-2.0
- Created
- Jul 23, 2026
- Updated
- Jul 23, 2026