yamatazen/Ayla-Light-16B

Partial fine-tune vs yamatazen/Ayla-Light-12B-Stock. Partial change: 45% of the weights changed (attention, mlp, norms), in 20 of 40 layers.

Tensors changed: lm_head 0/1, Embeddings 0/1, Norms 40/81, MLP 60/120, Attention 80/160

2 derivatives checked