SmolVLA: Efficient Vision-Language-Action Model trained on Lerobot Community Data

Look, they're pushing this efficiency narrative hard, but honestly, it's just fine-tuning a large model on curated data. SmolVLA isn't some revolutionary breakthrough; it's an optimization trick. We're seeing smaller models do more when the training data is focused, which is what the Lerobot community data gives it. It's practical optimization, not magic scaling.

We need to stop treating model size as the only metric. If the resulting action fidelity is high, the size is irrelevant. Don't overhype the distribution, just ship the working code.