OpenGVLab/InternVL2-Llama3-76B
Full fine-tune (fewer layers) vs NousResearch/Hermes-2-Theta-Llama-3-70B. NousResearch/Hermes-2-Theta-Llama-3-70B matches part of OpenGVLab/InternVL2-Llama3-76B (721 of its 1318 tensors; the rest of OpenGVLab/InternVL2-Llama3-76B has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 80 of 80 layers.
Tensors changed: Norms 160/161, MLP 240/240, Attention 320/320