CoolFace
Apppublic

ZeroxZhang/Qwen3.8-27B-Uncensored-MLX-Demo

sourceHugging Faceupdated 1mo agoView on Hugging Face
0likes
App README

Qwen3.8-27B-Uncensored-MLX Demo

An interactive demo page for the Qwen3.8-27B-Uncensored-MLX model — an abliterated MLX build of Qwen's 27B vision-language model for Apple Silicon.

Live Chat Demo

The embedded demo below uses the PyTorch equivalent on ZeroGPU. For native MLX inference on your Mac, see the local usage instructions.

Model Overview

Qwen3.8-27B-Uncensored-MLX is an abliterated (refusal-removed) MLX build of `Qwen/Qwen3.8-27B` — a 27B-parameter dense, hybrid-attention native vision-language model with thinking control, tool-calling and an MTP head.

Key Features

  • —Abliterated: Safety alignment removed for research purposes
  • —MLX Format: Optimized for Apple Silicon (M1/M2/M3/M4)
  • —Vision-Language: Native image understanding capabilities
  • —Multiple Quantizations: 2 / 4 / 6 / 8-bit precision options
  • —Long Context: 262K token context window

Available Quantizations

PrecisionSizeMin Mac RAMQuality
8-bit~27.5 GB32 GBNear-lossless
6-bit~22 GB24-32 GBExcellent
4-bit~15 GB24 GBVery good (recommended)
2-bit~8.7 GB16 GBSeverely degraded

Local Usage

bash
# Install mlx-vlm
pip install -U mlx-vlm

# Download the 4-bit quantization
hf download orcarouter/Qwen3.8-27B-Uncensored-MLX --include "4-bit/*" \
    --local-dir ./Qwen3.8-27B-Uncensored-MLX

# Text generation
python -m mlx_vlm generate \
    --model ./Qwen3.8-27B-Uncensored-MLX/4-bit \
    --prompt "Explain quantum entanglement in one sentence." --max-tokens 256

# Vision (image + text)
python -m mlx_vlm generate \
    --model ./Qwen3.8-27B-Uncensored-MLX/4-bit \
    --image path/to/image.png \
    --prompt "Describe this image." --max-tokens 256

# OpenAI-compatible server
python -m mlx_vlm server --model ./Qwen3.8-27B-Uncensored-MLX/4-bit --port 8080

Links

Disclaimer

This model has had its safety alignment substantially removed via abliteration. It is released strictly for legitimate research — interpretability, AI-safety and refusal-mechanism study, red-teaming, robustness evaluation, and controlled experiments. You assume full responsibility and liability for how you use it.