dphn/dolphin-2.9.3-mistral-nemo-12b
LoRA-style vs mistralai/Mistral-Nemo-Base-2407. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of down_proj, gate_proj, k_proj, o_proj, q_proj, up_proj, v_proj changed, in 40 of 40 layers; norms, biases and embeddings are untouched.
Tensors changed: Norms 0/81, MLP 60/120, Attention 80/160
3 derivatives checked
- win10/Mistral-Nemo-abliterated-Nemo-Pro-v2: Full fine-tune
- win10/ChatML-Nemo-Pro-V2: Full fine-tune
- HagalazAI/CyberDolphin-2.9.3-mistral-nemo-12b: LoRA-style
2 other repos hold byte-for-byte identical weight files.