All model and fine-tuning credit belongs to the original author, AEON-7, and the upstream Qwen team. This repository only provides an MLX conversion and quantization for easier local use on Apple Silicon. No additional fine-tuning or intentional behavioral changes were made.

Quantization

Format: MLX
Quantization: 4-bit affine
Group size: 64
Approximate download size: 14 GB
Intended platform: Apple Silicon macOS
Recommended unified memory: 32 GB or more

Install

pip install -U mlx-lm

Run an interactive chat

bash

mlx_lm.chat \
  --model choppedgarlic/Qwen3.8-27B-AEON-ULTIMATE-UNCENSORED-4bit-MLX \
  --max-tokens 8192 \
  --temp 0.6 \
  --top-p 0.95

The first launch downloads the model from Hugging Face. Later launches use the local Hugging Face cache.

Run an OpenAI-compatible local server

Thinking enabled:

bash

mlx_lm.server \
  --model choppedgarlic/Qwen3.8-27B-AEON-ULTIMATE-UNCENSORED-4bit-MLX \
  --host 127.0.0.1 \
  --port 8081 \
  --max-tokens 8192 \
  --temp 0.6 \
  --top-p 0.95 \
  --chat-template-args '{"enable_thinking":true}'

Thinking disabled:

bash

mlx_lm.server \
  --model choppedgarlic/Qwen3.8-27B-AEON-ULTIMATE-UNCENSORED-4bit-MLX \
  --host 127.0.0.1 \
  --port 8081 \
  --max-tokens 8192 \
  --temp 0.6 \
  --top-p 0.95 \
  --chat-template-args '{"enable_thinking":false}'

The API base URL is then http://127.0.0.1:8081/v1.

Verified locally

Text generation
Interactive mlx_lm.chat
Thinking and non-thinking modes
mlx_lm.server through its OpenAI-compatible API
Local use through the Pi coding agent

Limitations

This conversion is intended for Apple Silicon and requires MLX.
The current conversion contains no vision_config or vision preprocessor assets. Image input and image generation are not supported or claimed.
Quantization may reduce quality compared with the BF16 source model.
The model can produce inaccurate, unsafe, or objectionable output. Validate outputs before using them in production or high-stakes settings.
Please also read the original model card for its intended use, behavior, and limitations.

License and attribution

Released under the Apache License 2.0, matching the source repository's declared license. See LICENSE.

Source model: AEON-7/Qwen3.8-27B-AEON-ULTIMATE-UNCENSORED-BF16
Upstream family: Qwen
MLX conversion and quantization: choppedgarlic