Quantization backend(fbgemm) is installed automatically during pip install?

Hello.

I’ve been trying to do quantization using torch quant lib.
I read this contents that we should set a proper backend qconfig(e.g. fbgemm for x86).
In the official tutorial, there isn’t any guide about installing of the backend.
I thought the backend(fbgemm) is installed automatically during pip install (torch or something else), is this right?
Or, I should install the fbgemm on my linux server manually like this?(https://github.com/pytorch/FBGEMM)

Thanks in advance for your help.

Hi Je,

FBGEMM is a third party library of pytorch and should already be automatically installed when you install pytorch, so you don’t have to manually install it separately. You should be able to use pytorch quantization out of the box.

Best,
-Andrew

1 Like