🐋 Qwen Uncensored Live Web Demo & MLX Hub
Live GPU-Powered Uncensored AI Playground & Apple Silicon Hub for orcarouter/Qwen3.8-27B-Uncensored-MLX
System Prompt Presets
0.01 1.5
0.1 1
64 2048
🚀 Native Execution on macOS with Apple Silicon (M1 / M2 / M3 / M4)
The orcarouter/Qwen3.8-27B-Uncensored-MLX weights are natively optimized for Apple's unified memory architecture via MLX.
📊 Unified Memory (RAM) Requirements for 27B Model
| Quantization | Model Weight Size | Recommended Unified RAM | Target Hardware |
|---|---|---|---|
| 4-bit | ~15.5 GB | 24 GB – 36 GB+ | M2/M3/M4 Pro, M1 Max |
| 8-bit | ~28.0 GB | 48 GB – 64 GB+ | M1/M2/M3 Max, M4 Max |
| 2-bit | ~8.5 GB | 16 GB | Base M1/M2/M3/M4 Macs |
🔬 About Qwen Uncensored & Abliteration
- Weight Modification (Abliteration): Orthogonalized residual activations to eliminate refusal directions while preserving reasoning, coding, and mathematical capabilities.
- Live GPU Backend: Running real-time PyTorch inference on Hugging Face ZeroGPU.
- Apple Silicon MLX Weights: Available at
orcarouter/Qwen3.8-27B-Uncensored-MLX.