look at building tensorflow package

[eyberg]

a base install of tensorflow has quite a few shared libs and almost 1800 files - no reason for the end-user to deal with this if the base pkg is enough for most purposes (and might include necessary library base outside of normal python package)

https://github.com/nanovms/ops/issues/477

[eyberg]

https://nanovms.com/dev/tutorials/gpu-accelerated-computing-nanos-unikernels