sh2orc/Llama-3-Korean-12B
Full fine-tune (fewer layers) vs sh2orc/Llama-3-Korean-8B. sh2orc/Llama-3-Korean-8B matches part of sh2orc/Llama-3-Korean-12B (289 of its 435 tensors; the rest of sh2orc/Llama-3-Korean-12B has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 32 of 32 layers.
Tensors changed: Norms 46/65, MLP 96/96, Attention 128/128