huihui-ai/Huihui-Qwen3.8-27B-abliterated

Hugging Face
VISIONConcurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 16, 2026License:apache-2.0Architecture:Transformer0.1K Open Weights Featherless Exclusive Warm

Huihui-Qwen3.8-27B-abliterated is a 27 billion parameter uncensored version of the Qwen3.8-27B model, developed by huihui-ai. This model has undergone 'abliteration' to remove refusal behaviors, making it suitable for use cases requiring less restrictive content generation. It retains a 32768 token context length and is designed for experimental and research applications where content filtering is intentionally reduced.

Loading preview...

Overview

Huihui-Qwen3.8-27B-abliterated is a 27 billion parameter language model derived from Qwen/Qwen3.8-27B. Its primary distinction is the removal of refusal mechanisms through an 'abliteration' process, making it an uncensored variant. This proof-of-concept implementation aims to demonstrate how to modify LLM behavior regarding content refusal without using TransformerLens, specifically by retaining the first 15 layers without ablation.

Key Characteristics

  • Uncensored Output: Designed to generate content without typical refusal behaviors, offering greater flexibility for specific applications.
  • Abliteration Technique: Utilizes a method to modify model responses, detailed in the remove-refusals-with-transformers project.
  • Qwen3.8 Base: Built upon the Qwen3.8-27B architecture, inheriting its foundational capabilities.
  • Ollama Support: Directly available for use with Ollama, simplifying deployment for local inference.

Usage Warnings & Considerations

This model is explicitly noted for its reduced safety filtering, which means it may produce sensitive, controversial, or inappropriate content. It is not suitable for all audiences or public-facing applications and is recommended for research and experimental use in controlled environments. Users bear full responsibility for generated content and must ensure compliance with legal and ethical standards. Continuous monitoring and manual review of outputs are strongly advised.