DragonBophades/Inkling-Gutenberg-DPO
Full fine-tune vs thinkingmachines/Inkling. Fine-tune: 94% of the weights changed, across 66 of 66 layers.
Tensors changed: MLP 247/552, Norms 0/310, Attention 329/666, other 0/13, Embeddings 0/11
Full fine-tune vs thinkingmachines/Inkling. Fine-tune: 94% of the weights changed, across 66 of 66 layers.
Tensors changed: MLP 247/552, Norms 0/310, Attention 329/666, other 0/13, Embeddings 0/11