neopolita/Qwen3.8-Flash-Next-Saikei-v1.0-4bit-mlx
Abliterated vs mlx-community/Qwen3.8-Flash-Next-4bit. Looks like abliteration (refusal removal): only the output projections (down_proj) changed, in 36 of 36 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/393, Attention 0/1052, Norms 0/248, MLP 108/976, other 0/7
1 derivatives checked
- neopolita/Qwen3.8-Flash-Next-Saikei-v1.0-gguf: Quantization