huihui-ai/Huihui-Spark-X2.5-4B-abliterated
Abliterated vs XHToken/Spark-X2.5-4B. Looks like abliteration (refusal removal): only the output projections (down_proj, out_proj) changed, in 36 of 36 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/73, MLP 36/108, Attention 36/108
1 derivatives checked
- okenk/Huihui-Spark-X2.5-4B-abliterated-GGUF: Quantization