metadata
license: apache-2.0
library_name: mlx
tags:
- mlx
- int8
- automatic-speech-recognition
- speaker-diarization
base_model:
- vanch007/mlx-MOSS-Transcribe-Diarize
pipeline_tag: automatic-speech-recognition
mlx-MOSS-Transcribe-Diarize 8bit
8bit MLX weight-only quantized variant of vanch007/mlx-MOSS-Transcribe-Diarize.
Quantization uses MLX affine quantization with bits=8 and group_size=64 on the text backbone, excluding the Whisper audio encoder and VQ adaptor.
python -m moss_transcribe_diarize.mlx.cli /path/to/input.wav \
--model plaincompute/mlx-MOSS-Transcribe-Diarize-8bit \
--out-dir runs/mlx_8bit_example