louisbrulenaudet/Pearl-34B-ties

Full fine-tune vs abacusai/MetaMath-Bagel-DPO-34B. Fine-tune: 100% of the weights changed, across 60 of 60 layers.

Tensors changed: Embeddings 1/1, Norms 27/121, MLP 179/180, Attention 240/240, lm_head 1/1

2 derivatives checked