voidful/gemma-4-e4b-it-agent-sft-tw
Partial fine-tune vs google/gemma-4-E4B-it. Partial change: 58% of the weights changed (attention, embeddings, mlp, norms, other), in 42 of 42 layers.
Tensors changed: MLP 125/606, Norms 32/484, other 82/261, Attention 132/719, Embeddings 1/6