google/t5gemma-2b-2b-ul2-it
Partial fine-tune vs google/t5gemma-2b-2b-ul2. Partial change: 89% of the weights changed (attention, embeddings, mlp, norms), in 26 of 26 layers.
Tensors changed: Embeddings 1/2, MLP 156/156, Norms 262/262, Attention 312/312