Qwen3.8-27B-Uncensored-MLX: Uncensored Qwen3.8-27B for Local Inference on Apple Silicon

This model, built on the Qwen3.8-27B model, is optimized for local inference on Apple Silicon using the MLX framework. It features a multimodal architecture capable of processing text and images simultaneously and is distributed under the Apache 2.0 license.

Unlike standard LLMs, this 'Uncensored' version has safety filters removed, making it suitable for AI red teaming and unrestricted generation tasks. It offers various quantization options from 2-bit to 8-bit, allowing memory usage to be adjusted according to hardware specifications.

The model supports Function Calling and Reasoning through internal templates. Reasoning intensity can be set to three levels: xhigh, medium, and low, enabling flexible control based on use cases—such as requiring detailed thought processes for complex problem-solving or obtaining quick responses for simple queries.

This is useful for developers who need to process private data on their own hardware without relying on cloud APIs, or who wish to remove generation restrictions on specific topics. It serves as an alternative for running high-performance LLMs locally on Apple devices when existing models' response limitations are burdensome.