Theros/Q2.5-ColdBrew-R1-Atrax

Partial fine-tune vs SvalTek/Q2.5-ColdBrew-R1-Indigo. Partial change: 3% of the weights changed (attention, mlp, norms), in 1 of 35 layers.

Tensors changed: lm_head 0/1, Embeddings 0/1, Norms 2/71, MLP 3/105, Attention 7/245

1 derivatives checked

On the Hub · Downloads on Model Pulse