Downloads · 30 days
0
UMI-Benchmark-v1-checkpoints
UMIbenchmark/UMI-Benchmark-v1-checkpoints
UMI-Benchmark-v1-checkpoints is a machine learning model from UMIbenchmark. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
This private staging package contains one checkpoint per task for each supported model family.
Access
Public
Updated Jun 5, 2026
Repo size
635 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
Other633 GB · 100%
At a glance
- Access
- Public
- Created
- Jun 5, 2026
- Updated
- Jun 5, 2026
- SHA
- 2731c27e
Model card
From the Hugging Face model README
UMI-Benchmark v1 Checkpoints
This private staging package contains one checkpoint per task for each supported model family.
Models:
- OpenPI pi0
- OpenPI pi0.5
- DreamZero
Tasks:
dual_arm_task1dual_arm_task2dual_arm_task3dual_arm_task4dual_arm_task5dual_arm_task6single_arm_task1single_arm_task2single_arm_task3single_arm_task4
OpenPI config names are normalized as pi0_<task> and pi05_<task>.
OpenPI checkpoint paths use public step 60000.
DreamZero checkpoint metadata records the exported training step. Single-arm DreamZero checkpoints use 40000 steps, except single_arm_task4 which uses 20000 steps.
Checkpoints trained with 4-card and 8-card setups are equivalent for release use; the distinction only reflects the training parallelism/resource configuration.
The release manifest is in metadata/checkpoint_manifest.json.
- Created
- Jun 5, 2026
- Updated
- Jun 5, 2026