abacusai/Smaug-72B-v0.1
LoRA-style vs moreh/MoMo-72B-lora-1.8.7-DPO. 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 80 of 80 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/1, Norms 0/161, MLP 240/240, Attention 320/640, lm_head 0/1
2 derivatives checked
- saucam/Pyrhea-72B: LoRA-style (fewer layers)
- Syed-Hasan-8503/Exodius-70B: Full fine-tune (fewer layers)