See the repo here.
Also check the associate repo as recipe for conda build and package upload via CI. Numkl feedstock is built based on conda smithy.
Try the package by
conda install -c refraction-ray numkl, only linux is supported.
This package provide memory effcient 64-bit int interface for eigen decompositions backend by mkl’s routines
It is worth noting that 64-bit compatibility has been improved in numpy after the release of this package.
See this numpy commit for more information, the origin of the problem and for the package is my issue here. However numpy’s recent fix is not friendly enough for mkl instead of default openblas. The core of the problem lies in the name convention of symbols using 64 bit int. Therefore this package is still a ready-to-go solution to utilize MKL in exact diagonalizations of large size in python.