Robotics (VLA) · Hugging Face LeRobot
SmolVLA
450M vision-language-action policy that trains and runs on one consumer GPU.
Datasheet
Maker | Hugging Face LeRobot |
|---|---|
Parameters | 450 MVerifiedsafetensors.total from the HF API Source |
Licence | Apache-2.0Verified |
Access | Public download |
Task (HF pipeline) | robotics |
First published on HF | 2025-06-01 |
Last updated on HF | 2026-10-06 |
Weights in memory | 0.9 GB at 16-bit · 0.3 GB at ~4.85 bitsEstimateParameters × bits ÷ 8. Excludes context cache and runtime overhead; not every model has a 4-bit build. |
Benchmark
87.3%
LIBERO, average success rate (4 suites) · #1 of 3 scored in this category
SmolVLA (0.45B) row. Source
Used in
The description summarises the model card at https://huggingface.co/lerobot/smolvla_base. Parameters and licence are read from the Hugging Face API; the weights-in-memory line is our arithmetic, not a measurement.