djuna/G2-BigGSHT-27B-2

Full fine-tune vs AALF/gemma-2-27b-it-SimPO-37K-100steps. Fine-tune: 100% of the weights changed, across 46 of 46 layers.

Tensors changed: Embeddings 1/1, Norms 20/185, MLP 138/138, Attention 184/184

2 derivatives checked