Crata/GRM-3.2-Sky-abliterated
LoRA-style vs OrionLLM/GRM-3.2-Sky. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of down_proj, k_proj, o_proj, out_proj, q_proj, v_proj changed, in 36 of 40 layers; norms, biases and embeddings are untouched.
Tensors changed: lm_head 0/1, Embeddings 0/4, Norms 0/241, Attention 45/388, MLP 70/2366
2 derivatives checked
- mradermacher/GRM-3.2-Sky-abliterated-i1-GGUF: Quantization
- mradermacher/GRM-3.2-Sky-abliterated-GGUF: Quantization