huihui-ai/Huihui-Qwen3.5-4B-Claude-4.6-Opus-abliterated
Abliterated vs Jackrong/Qwen3.5-4B-Claude-4.6-Opus-Reasoning-Distilled. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj, out_proj) changed, in 32 of 32 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/5, MLP 33/199, Attention 33/324, Norms 0/209, other 0/1
6 derivatives checked
- mradermacher/Huihui-Qwen3.5-4B-Claude-4.6-Opus-abliterated-GGUF: Same weights, converted
- mradermacher/Huihui-Qwen3.5-4B-Claude-4.6-Opus-abliterated-i1-GGUF: Quantization
- zapabobouj/huihui-qwen35-4b-roleplay-qlora-ms2048-s800: LoRA-style
- zapabobouj/qwen3.5-elt-l3: LoRA-style (fewer layers)
- hotdogs/Qwen3.5-4B-MoLE: LoRA-style
- agrawal-prateek/qwen3.5-4b-abliterated-gguf: Same weights, converted
2 other repos hold byte-for-byte identical weight files.