---

library_name: transformers

license: apache-2.0

pipeline_tag: image-text-to-text

base_model:

• nerkyor/Qwen3.8-27B-Coder390-EfficientThink-Opus5.5-GPT6Astra-Grok4.7-DSV4Pro-K3-SFT-RLOO-MTP-DFlash2

tags:

• abliterated
• uncensored
• huihui
• qwen3
• imatrix

---

huihui-ai/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated-GGUF

This is an uncensored version of nerkyor/Qwen3.8-27B-Coder390-EfficientThink-Opus5.5-GPT6Astra-Grok4.7-DSV4Pro-K3-SFT-RLOO-MTP-DFlash2 created with abliteration (see remove-refusals-with-transformers to know more about it).

This is a crude, proof-of-concept implementation to remove refusals from an LLM model without using TransformerLens.

Note

Only layers 2 to 52 have been ablated.

llama.cpp

Use the latest llama.cpp,

llama-cli -m huihui-ai/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated-GGUF/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated-IQ3_S3.gguf -c 262144

GGUF

1. download llama.cpp-b11540

1. download GGUF

hf download huihui-ai/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated-GGUF --local-dir ./huihui-ai/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated-GGUF

1. llama-imatrix

calibration1.txt, calibration2.txt, calibration3.txt come from eaddario/imatrix-calibration

llama-imatrix -m huihui-ai/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated-GGUF/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated-bf16.gguf -f huihui-ai/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated-GGUF/calibration3.txt -o huihui-ai/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated-GGUF/abliterated-imatrix3.gguf --chunk 512 --parse-special -b 2048 --no-ppl

1. llama-quantize

llama-quantize --allow-requantize --tensor-type-file huihui-ai/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated-GGUF/tensor-type-allocation-IQ3_S.txt --imatrix huihui-ai/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated-GGUF/abliterated-imatrix3.gguf huihui-ai/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated-GGUF/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated-bf16.gguf huihui-ai/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated-GGUF/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated-IQ3_S3.gguf Q3_K

1. llama-perplexity

llama-perplexity -m huihui-ai/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated-GGUF/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated-IQ3_S3.gguf -c 8192 -b 2048 -ctk bf16 -ctv bf16 -f huihui-ai/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated-GGUF/calibration2.txt

1. llama-perplexity (calibration1.txt)

model | calibration | perplexity | PPL | Gap |

-----------------------------------------------------------------------------|------------------|------------------|--------------------------|-------------------|

nerkyor/Qwen3.8-27B-Coder390-EfficientThink.gguf | | calibration1.txt | PPL = 4.1744 +/- 0.00925 | + 0 |

huihui-ai/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated-bf16.gguf | | calibration1.txt | PPL = 4.1588 +/- 0.00916 | - 0.0156 + 0 |

huihui-ai/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated-IQ3_S1.gguf | calibration1.txt | calibration1.txt | PPL = 4.1922 +/- 0.00928 | + 0.0178 + 0.0334 |

huihui-ai/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated-IQ3_S2.gguf | calibration2.txt | calibration1.txt | PPL = 4.1864 +/- 0.00926 | + 0.0120 + 0.0276 |

huihui-ai/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated-IQ3_S3.gguf | calibration3.txt | calibration1.txt | PPL = 4.1851 +/- 0.00925 | + 0.0107 + 0.0263 |

1. llama-perplexity (calibration2.txt)

model | calibration | perplexity | PPL | Gap |

-----------------------------------------------------------------------------|------------------|------------------|--------------------------|-------------------|

nerkyor/Qwen3.8-27B-Coder390-EfficientThink.gguf | | calibration2.txt | PPL = 3.8816 +/- 0.00677 | + 0 |

huihui-ai/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated-bf16.gguf | | calibration2.txt | PPL = 3.8468 +/- 0.00664 | - 0.0348 + 0 |

huihui-ai/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated-IQ3_S1.gguf | calibration1.txt | calibration2.txt | PPL = 3.8547 +/- 0.00665 | - 0.0269 + 0.0079 |

huihui-ai/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated-IQ3_S2.gguf | calibration2.txt | calibration2.txt | PPL = 3.8495 +/- 0.00664 | - 0.0321 + 0.0027 |

huihui-ai/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated-IQ3_S3.gguf | calibration3.txt | calibration2.txt | PPL = 3.8458 +/- 0.00662 | - 0.0358 - 0.0010 |

1. calibration datasets

name | imatrix | size | records |

------------------|---------------------------|------------|---------|

calibration1.txt | abliterated-imatrix1.gguf | 8,139,994 | 18,749 |

calibration2.txt | abliterated-imatrix2.gguf | 12,760,564 | 24,999 |

calibration3.txt | abliterated-imatrix3.gguf | 22,079,008 | 37,449 |

Usage Warnings

- Risk of Sensitive or Controversial Outputs: This model’s safety filtering has been significantly reduced, potentially generating sensitive, controversial, or inappropriate content. Users should exercise caution and rigorously review generated outputs.

- Not Suitable for All Audiences: Due to limited content filtering, the model’s outputs may be inappropriate for public settings, underage users, or applications requiring high security.

- Legal and Ethical Responsibilities: Users must ensure their usage complies with local laws and ethical standards. Generated content may carry legal or ethical risks, and users are solely responsible for any consequences.

- Research and Experimental Use: It is recommended to use this model for research, testing, or controlled environments, avoiding direct use in production or public-facing commercial applications.

- Monitoring and Review Recommendations: Users are strongly advised to monitor model outputs in real-time and conduct manual reviews when necessary to prevent the dissemination of inappropriate content.

- No Default Safety Guarantees: Unlike standard models, this model has not undergone rigorous safety optimization. huihui.ai bears no responsibility for any consequences arising from its use.

Donation

Your donation helps us continue our further development and improvement, a cup of coffee can do it.

• bitcoin:

  bc1qqnkhuchxw0zqjh2ku3lu4hq45hc6gy84uk70ge

• Support our work on Ko-fi!