jmd1011/snek-LMS

Open Source z3py_overloads

Closed this issue · 1 comments

Hi,

I'm a pytorch developer looking to implement something similar or build off the of the work you've done in LMS staging. In particular, I am trying to interpret through shape formulas written in python in order to build constraints of tensor shape logic.

I've previously had some success partially evaluating shape logic and then lowering it to z3 and I would also like to explore similar lowering through LMS.

I read in your paper a description of z3py_overloads - would it be possible for you to open source this or otherwise help share your implementation ?

Thank you for your work here!

I was pointed to https://github.com/google/pyctr! Closing