Run the 8-bit model
Use an Apple Silicon Mac and a native ARM64 Python environment. This command loads the named MLX checkpoint and transcribes the sample locally.
macOS terminal
python -m pip install mlx-audio
curl --fail --location --output sample.wav https://oruk.ai/samples/oruk-quickstart.wav
python -m mlx_audio.stt.generate --model mlx-community/Qwen3-ASR-1.7B-8bit --audio sample.wavCompare a smaller conversion
The community also publishes a 4-bit conversion of the Qwen base. Compare the transcript and memory use on the same file.
4-bit base model
python -m mlx_audio.stt.generate --model mlx-community/Qwen3-ASR-1.7B-4bit --audio sample.wav