Blackfrost-AI/Qwen3.8-27B-ABLITERATED-BF16
Abliterated vs Qwen/Qwen3.8-27B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj, out_proj) changed, in 62 of 64 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Norms 0/325, Attention 62/560, MLP 62/307, Embeddings 0/5, lm_head 0/1, other 0/1
26 derivatives checked
- Blackfrost-AI/Qwen3.8-27B-ABLITERATED-GGUF: Not comparable yet
- Blackfrost-AI/Qwen3.8-27B-ABLITERATED-NVFP4: Quantization
- davetha/Qwen3.8-27B-abliterated-W8A16: Quantization
- mradermacher/Qwen3.8-27B-ABLITERATED-BF16-i1-GGUF: Quantization
- 1105s110/Qwen3.8-27B-Blackfrost-Abliterated-GSQ-RCO-IQ3_XXS-GGUF: Quantization
- qzshch/Qwen3.8-27B-Blackfrost-Abliterated-NVFP4-GGUF: Quantization
- mradermacher/Qwen3.8-27B-ABLITERATED-BF16-GGUF: Quantization
- grant-ai/Qwen3.8-27B-Abliterated-MTPLX-4bit: Quantization
- EgorKodin/Qwen3.8-27B-ABLITERATED-3bit-MLX-TextOnly: Quantization
- grant-ai/Qwen3.8-27B-Abliterated-MTPLX: Quantization
- Ishowbackup/Qwen3.8-27B-ABLITERATED-GGUF: Not comparable yet
- Verkiki/Qwen3.8-27b-Fable5-Distill-Abliterated-GGUF: Not comparable yet
- cvgro/Qwen3.8-27B-ABLITERATED-GGUF: Quantization
- EgorKodin/Qwen3.8-27B-ABLITERATED-2bit-MLX-TextOnly: Quantization
- grant-ai/Qwen3.8-27B-Abliterated-MTPLX-6bit: Quantization
- Xeldia/Qwen3.8-27B-ABLITERATED-GGUF: Not comparable yet
- asdasdadadsss/Qwen3.8-27B-ABLITERATED-GGUF: Not comparable yet
- 1105s110/Qwen3.8-27B-Blackfrost-Abliterated-SlotTuned-15G-GGUF: Quantization
- lyf/Qwen3.8-27B-Blackfrost-Abliterated-NVFP4-MTP-VL: Quantization
- ap3212/Qwen3.8-27B-ABLITERATED-3bit-MLX-TextOnly: Quantization
- wfiedler/Qwen3.8-27B-ABLITERATED-mlx-q3km: Quantization
- WatchDG/Qwen3.8-27B-ABLITERATED-EXL3-3.75bpw: Same weights, converted
- WatchDG/Qwen3.8-27B-ABLITERATED-EXL3-2.75bpw: Same weights, converted
- davetha/Qwen3.8-27B-ABLITERATED-W8A8-gdnint8: Quantization
- davetha/Qwen3.8-27B-abliterated-W8A8: Quantization
- WatchDG/Qwen3.8-27B-ABLITERATED-EXL3-2.25bpw: Same weights, converted
1 other repos hold byte-for-byte identical weight files.