PocketAI Model Lab has released a family of unofficial MLX derivatives of the Qwen/Qwen3.8-27B model. These variants, including 2-bit AWQ, 4-bit, 6-bit, 8-bit, and BF16 versions, have been modified to suppress learned refusal behavior, potentially leading to more harmful or offensive content. The models have undergone behavioral evaluation and feature validation, with most variants passing quality and tool-call checks, though the 2-bit AWQ build is noted as experimental with limitations. AI
IMPACT These modified models may offer new capabilities but also introduce risks due to suppressed refusal behavior, requiring careful evaluation by users.
RANK_REASON Release of modified open-source models with a focus on altered safety behaviors. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Hugging Face Trending Models →
- Atomic Chat
- Google Colab
- Kaggle
- LM Studio
- MLX
- mlx-vlm
- PocketAiHub/Qwen3.8-27B-Abliterated-MLX
- PocketAI Model Lab
- Qwen
- Qwen/Qwen3.8-27B
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
