HuggingFaceVLA/smolvla_libero

Partial fine-tune vs lerobot/smolvla_base. Partial change: 1% of the weights changed (attention, other), in 0 of 16 layers.

Tensors changed: other 3/4, Attention 16/176, lm_head 0/1, Embeddings 0/4, Norms 0/83, MLP 0/96

1 derivatives checked