Downloads · 30 days
0
action-tokenizer-axis2-matched-noreg-split-100k-slurm-20393-20260923
junbrro/action-tokenizer-axis2-matched-noreg-split-100k-slurm-20393-20260923
action-tokenizer-axis2-matched-noreg-split-100k-slurm-20393-20260923 is a machine learning model from junbrro. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
Final step: 100000. Source: slurm-20393.
Access
Public
Updated Sep 23, 2026
Repo size
77.2 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pt38.6 MB · 48%
At a glance
- Access
- Public
- Created
- Sep 23, 2026
- Updated
- Sep 23, 2026
- SHA
- e5f4628a
Model card
From the Hugging Face model README
Arm 2 matched action-only tokenizer without embodiment regularization
Final step: 100000. Source: slurm-20393.
Final model weights and configuration only; optimizer/RNG state excluded. Original configuration is preserved, including source-cluster paths. Remap paths before use. Bundled actlat/ contains the action tokenizer when applicable.
No embodiment regression/VICReg; split robot/human reconstruction decoders, L1 reconstruction and KL 1e-6. Use bundle/ for VLA. Normalization is preserved exactly. Robot BBT + Human Open/Close. This is the independently trained Slurm tokenizer; VLA completion is a separate stage.
- Created
- Sep 23, 2026
- Updated
- Sep 23, 2026