https://github.com/cassiebreviu/whisper_quant

https://github.com/cassiebreviu/whisper_quant

Last synced: 24 days ago
JSON representation

• Host: GitHub
• URL: https://github.com/cassiebreviu/whisper_quant
• Owner: cassiebreviu
• Created: 2023-04-10T23:34:48.000Z (over 1 year ago)
• Default Branch: main
• Last Pushed: 2023-04-08T06:05:35.000Z (over 1 year ago)
• Last Synced: 2024-04-23T14:30:39.541Z (9 months ago)
• Size: 2.15 MB
• Stars: 1
• Watchers: 0
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.txt

Awesome Lists containing this project

README

        
1. install conda env:

    $ conda env create -f whisper.yml

    $ conda activate whisper

2. clone and install optimum:

    $ git clone --recursive https://github.com/yufenglee/optimum

    $ cd optimum

    $ git checkout whisper_static_quant

    $ python -m pip install -e .

3. insall onnxruntime:

    $ pip install onnxruntime

4. go to your work directory and download and save fp32 whisper model:

    

    $ python save_model.py

5. install ffmpeg

    $ conda install ffmpeg

6. quantize model with "quant.sh"

    $ bash quant.sh

7. run and verify the accuracy:

    $ python simple_audio_wer.py