mayacinka/djinn
Full fine-tune vs CultriX/NeuralTrix-7B-dpo. Fine-tune: 97% of the weights changed, across 31 of 32 layers.
Tensors changed: Norms 62/65, MLP 93/96, Attention 124/128
Full fine-tune vs CultriX/NeuralTrix-7B-dpo. Fine-tune: 97% of the weights changed, across 31 of 32 layers.
Tensors changed: Norms 62/65, MLP 93/96, Attention 124/128