寒武纪MLU 适配faster whisper

适配过程

# 使用寒武纪官网下载的pytorch镜像起容器，进入容器后
# git代码
git clone https://github.com/SYSTRAN/faster-whisper.git
# 模型转换
cd /whisper
python /torch/src/catch/tools/torch_gpu2mlu/torch_gpu2mlu.py -i faster_whisper/
# 获得faster_whisper_mlu代码文件夹

下载asteroid-filterbanks代码

git clone https://github.com/asteroid-team/asteroid-filterbanks.git

python /torch/src/catch/tools/torch_gpu2mlu/torch_gpu2mlu.py -i asteroid-filterbanks/

cd astroid-filterbanks_mlu
# 注释掉requirements.txt中的# torch>=1.8.0
pip install -e .

下载与torch对应版本的lightning

对应版本可以查看lightning官网Versioning Policy — PyTorch Lightning 2.4.0 documentation 进行查看

pip install pytorch-lightning==2.3.0