Goekdeniz-Guelmez/Josiefied-Qwen3-1.7B-abliterated-v1
Abliterated vs Qwen/Qwen3-1.7B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 27 of 28 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/113, MLP 27/84, Attention 27/112
6 derivatives checked
- mlx-community/Josiefied-Qwen3-1.7B-abliterated-v1-4bit: Quantization
- Goekdeniz-Guelmez/Josiefied-Qwen3-1.7B-abliterated-v1-gguf: Same weights, converted
- mradermacher/Josiefied-Qwen3-1.7B-abliterated-v1-GGUF: Same weights, converted
- DavidAU/Qwen3-MOE-6x1.7B-10.2B-Shining-Madness-Uncensored: MoE built from dense models
- mlx-community/Josiefied-Qwen3-1.7B-abliterated-v1-bf16: Same weights, repacked
- DavidAU/Qwen3-MOE-4x1.7B-6.8B-Lucys-Song-Uncensored: MoE built from dense models