google/t5gemma-ml-ml-ul2-it
Partial fine-tune vs google/t5gemma-ml-ml-ul2. Partial change: 87% of the weights changed (attention, embeddings, mlp, norms), in 26 of 26 layers.
Tensors changed: Embeddings 1/2, Attention 312/312, MLP 156/156, Norms 262/262