1.8k likes
Uncensored Qwen3.8 27B model stripped of refusal filters
What is Huihui-Qwen3.8-27B-abliterated-GGUF?
This GGUF model is a 27-billion-parameter version of Qwen3.8 that has had its refusal-handling layers removed, so it will answer prompts without the usual safety blocks. The ablation targets layers 22-52 and mixes weights from Swift-1.5 and Underdog-Saluki, released as a proof-of-concept on Hugging Face.
Who it is for
If you need to see how a large language model behaves without built-in refusal logic, you can download and run this model locally.
How to try it
Download the GGUF file from the repo and run it with llama.cpp, e.g. `./main -m model.gguf`.