Lambent/Mira-v1.20-27B-dpo
LoRA-style vs Lambent/Mira-v1.19.3c-27B-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 62 of 62 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/4, Norms 0/484, MLP 186/294, Attention 248/464, other 0/1
4 derivatives checked
- mradermacher/Mira-v1.20-27B-dpo-GGUF: Quantization (fewer layers)
- mradermacher/Mira-v1.20-27B-dpo-i1-GGUF: Quantization (fewer layers)
- Lambent/Mira-v1.21-27B-rlvr: LoRA-style
- Lambent/Mira-v1.24.2-27B-Karcher: Full fine-tune