Sakalti/ultiima-14B-v0.3

Partial fine-tune vs sometimesanotion/Qwen2.5-14B-Vimarckoso-v3. Partial change: 5% of the weights changed (attention, mlp, norms), in 20 of 48 layers.

Tensors changed: lm_head 0/1, Embeddings 0/1, Norms 5/97, MLP 7/144, Attention 11/336

1 derivatives checked