websfactory/Webs-Sejong-31B-v7

Full fine-tune vs google/gemma-4-31B-it. Fine-tune: 94% of the weights changed, across 60 of 60 layers.

Tensors changed: Embeddings 0/4, Norms 120/523, other 0/62, MLP 180/261, Attention 230/338

2 derivatives checked