cadazar/han2han-rl
Partial fine-tune vs cadazar/han2han-it. Partial change: 90% of the weights changed (attention, embeddings, mlp, norms, other), in 18 of 18 layers.
Tensors changed: Norms 292/292, Attention 360/360, MLP 162/162, other 6/10, Embeddings 1/1