---

一、paas平台环境选择

驱动选择：5.10.22及以上

镜像选择：pytorch2.1

二、代码环境准备

1.代码下载

git clone https://github.com/jianchang512/chatTTS-ui
python /torch/src/catch/tools/torch_gpu2mlu/torch_gpu2mlu.py -i chatTTS-ui/

2.环境安装

modelsope

git clone https://github.com/modelscope/modelscope.git
python /torch/src/catch/tools/torch_gpu2mlu/torch_gpu2mlu.py -i modelscope/
pip install -r modelscope/requirements/framework.txt
pip install -e ./modelscope_mlu/

transformers

git clone -b v4.39.2 https://githubfast.com/huggingface/transformers.git
pip install -e ./transformers/

accelerate

git clone https://githubfast.com/huggingface/accelerate.git
pip install -e ./accelerate/

3.常规pip安装

cd chatTTS-ui_mlu && pip install -r requirements.txt

4.代码修改

vocos/heads 

 	+	mag = mag.cpu()
    +   x = x.cpu()
    +   y = y.cpu()
        S = mag * (x + 1j * y)
    +    mag = mag.mlu()
    +    x = x.mlu()
    +   y = y.mlu()
    +    S =S.mlu()

4.代码修改

三.算法启动

模型默认采用魔塔路径下载，这里不做模型下载教学

python app.py

MLU体验感拉满了简直
、