mergekit-community/mergekit-slerp-crxrtap
Full fine-tune vs beomi/gemma-ko-2b. Fine-tune: 96% of the weights changed, across 18 of 18 layers.
Tensors changed: Embeddings 1/1, Norms 37/37, MLP 51/54, Attention 68/72
Full fine-tune vs beomi/gemma-ko-2b. Fine-tune: 96% of the weights changed, across 18 of 18 layers.
Tensors changed: Embeddings 1/1, Norms 37/37, MLP 51/54, Attention 68/72