jadanovitch/t5gemma-2-4b-encoder
Partial fine-tune vs google/t5gemma-2-4b-4b. Partial change: 83% of the weights changed (attention, mlp, norms), in 34 of 34 layers.
Tensors changed: Embeddings 0/1, MLP 102/102, Norms 136/137, Attention 136/136
Partial fine-tune vs google/t5gemma-2-4b-4b. Partial change: 83% of the weights changed (attention, mlp, norms), in 34 of 34 layers.
Tensors changed: Embeddings 0/1, MLP 102/102, Norms 136/137, Attention 136/136