Lambent/Mira-v1.9-27B

Full fine-tune vs Lambent/Mira-v1.8.1a-27B. Fine-tune: 93% of the weights changed, across 62 of 62 layers.

Tensors changed: Embeddings 0/4, Norms 11/484, MLP 194/294, Attention 260/464, other 0/1

3 derivatives checked