Downloads · 30 days
2
7% of all-time downloads
eval3-smolvla-3way-v5-newdata-balanced
RobotLearningVLA/eval3-smolvla-3way-v5-newdata-balanced
eval3-smolvla-3way-v5-newdata-balanced is a machine learning model from RobotLearningVLA. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
All-time downloads
27
Public
Parameters
450M
907 MB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors907 MB · 100%
Parameter types
How the weights are stored.
BF16447M · 99%
Model card
From the Hugging Face model README
Eval3 SmolVLA 3-Way v5 New Data Balanced
Selected checkpoint: 050000.
This checkpoint was trained fresh from lerobot/smolvla_base on the new Eval 3 datasets only:
Taylor Swift, Barack Obama, and Yann LeCun across left, middle, and right target positions.
Selection
The 050000 checkpoint was selected over 060000 by offline checkpoint sweep:
- MAE:
1.1335 - Final-phase MAE:
1.3801 - Prompt-nearest accuracy:
59.1% - Final prompt accuracy:
68.2% - Normalizer count:
59933for bothactionandobservation.state
Limitations
This is not a guarantee of physical success. Hardware validation is still required. OOD celebrity behavior is not solved by this checkpoint.
- Created
- May 16, 2026
- Updated
- May 16, 2026