make the threshold for acurracy more precise (#17194)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/17194
we found that there is a per row absolute error due to int8 quant
and a relative error table-wide in case fp16 is used
Reviewed By: csummersea
Differential Revision: D14113353
fbshipit-source-id: c7065aa9d15c453c2e5609f421ad0155145af889