yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2
Full fine-tune vs google/gemma-4-12B-it. Fine-tune: 91% of the weights changed, across 48 of 48 layers.
Tensors changed: Embeddings 0/12, Norms 285/289, other 6/48, MLP 144/144, Attention 184/184
2 derivatives checked
- huihui-ai/Huihui-gemma-4-12B-agentic-fable5-abliterated: Abliterated
- droplychee-core/droplychee-flash-v2: LoRA-style
1 other repos hold byte-for-byte identical weight files.