CamilleWalters/smolvla_full-block-sorter-v1
Partial fine-tune vs lerobot/smolvla_base. Partial change: 77% of the weights changed (attention, embeddings, mlp, norms, other), in 15 of 16 layers.
Tensors changed: other 7/7, MLP 145/148, Attention 220/224, Norms 43/116, lm_head 0/1, Embeddings 3/4