Martiño Crespo

I also tried to use the base model without fine-tuning it. More specifically, I tried to evaluate a very similar task to this one: https://huggingface.co/spaces/lerobot/visualize_dataset?path=%2Flerobot%2Fsvla_so101_pickplace%2Fepisode_0.

Since the provided model is trained on that specific data, I would expect to successfully reproduce the completion of a very similar task, even with my own setup and objects.

However, it was a complete failure. Could you please be more specific about this statement?

Currently, the LeRobot library doesn’t natively support zero-shot inference with the pre-trained SmolVLA model. It requires access to the action and feature normalization stats used during training, which aren’t yet exposed in the config. There’s some discussion around enabling this functionality in the future (which I’m very excited about and will definitely keep an eye on!).