4-bit MLX quantization of
philbert440/Qwen3.8-27B-Uncensored-Cyber
for
Apple Silicon (MLX). Group size 64, affine 4-bit.
Cyber/offensive-security specialized de-refusal of Qwen3.8-27B (built on the Aggressive base + a residual-cyber
peel) — fully open on the cyber domain. See the base model card for recipe and evaluation details.
1pip install mlx-lm
2mlx_lm.generate --model philbert440/Qwen3.8-27B-Uncensored-Cyber-MLX-4bit --prompt "..."
Uncensored model — you are responsible for how you use it. Thinking-mode use benefits from an ample token
budget and sampling (temp ~0.7); greedy/temperature-0 can loop on long open-ended prompts.
