yanolja/YanoljaNEXT-EEVE-Instruct-10.8B
Full fine-tune vs yanolja/YanoljaNEXT-EEVE-10.8B. Fine-tune: 98% of the weights changed, across 48 of 48 layers.
Tensors changed: Embeddings 0/1, Norms 0/97, MLP 144/144, Attention 192/192, lm_head 1/1
3 derivatives checked
- maywell/Jolteon-Instruct-13B-alpha: Full fine-tune
- Suchae/EEVE-Korean-Judgment-Transducer-10.8B-v2.0: LoRA-style
- ENERGY-DRINK-LOVE/eeve_dpo-v3: Full fine-tune