allenai/Molmo2-ER
Partial fine-tune vs allenai/Molmo2-4B. Partial change: 84% of the weights changed (attention, embeddings, mlp, norms, other), in 36 of 36 layers.
Tensors changed: Norms 245/245, MLP 172/172, Attention 272/272, Embeddings 3/5, lm_head 0/1, other 11/11