Modelpublic

FredrikKarlssonSpeech/roest-v3-whisper-1.5b-mlx-4bit

sourceHugging Faceopenrailupdated 25d agoView on Hugging Face

0likes125downloads

Model Card

roest-v3-whisper-1.5b (MLX, 4-bit quantized)

MLX conversion of CoRal-project/roest-v3-whisper-1.5b for fast inference on Apple Silicon via mlx-whisper.

Converted with mlx-examples/whisper/convert.py at 4-bit quantized precision.

Usage

bash

pip install mlx-whisper
mlx_whisper audio_file.mp3 --model FredrikKarlssonSpeech/roest-v3-whisper-1.5b-mlx-4bit

python

import mlx_whisper
result = mlx_whisper.transcribe("audio_file.mp3", path_or_hf_repo="FredrikKarlssonSpeech/roest-v3-whisper-1.5b-mlx-4bit")
print(result["text"])