Huihui Qwen3.8 27B Abliterated

About Huihui Qwen3.8 27B Abliterated

Huihui Qwen3.8 27B Abliterated is a community-modified version of Alibaba's Qwen3.8-27B open-weight model, published on Hugging Face by the huihui-ai account as GGUF files for local inference. The modification, called abliteration, suppresses the refusal behaviour of the original model by editing a subset of its layers, while the vision components and other capabilities are left in place. The repository offers quantizations from roughly 7 GB to 55 GB and works with llama.cpp-based software such as Ollama, LM Studio and KoboldCpp.

The model has 27 billion parameters and a 262,144-token context window. The base model is released under the Apache-2.0 license, and the same license appears in the repository metadata. It is aimed at users with a 16-24 GB GPU or a Mac with 32 GB of unified memory who want an unrestricted general-purpose assistant that runs entirely on their own machine. The model card describes the abliteration method as an experimental, proof-of-concept technique, so output quality can differ from the unmodified base model.