Add Required library optimization for CPU VS GPU

[DevanshMistry890]

I tested that while running locally on the CPU, package install cuda GPU libraries. It is taking up too much space for that additional libraries. Since my CPU will not go to use. So add an option for installing libraries specific to chosen CPU or GPU only. This will also help in terms of memory optimization while deployment on docker.

[yangheng95]

It's weired cuz it you direct install pyabsa from pip, then it usually install the cpu version. Do you mean it downloads the cuda-included torch lib? Can you shou some clues to help me resolve this issue?

[DevanshMistry890]

I am not perfectly aware of all your libraries but you can read this article and try to run time optimizing torch library. optimization I know the task is very tedious you can also tag this issue as a recommendation.

[yangheng95]

Thanks for your advice, I will check what I can do next.

[yangheng95]

For anyone who have the similar problem, there is a temporary method: please install the torch depends on your situation before you install PYABSA.