Daniel Smith
Daniel Smith
@janeyx99 I'm looking to push another release shortly, is this still a requested feature?
I'll get in #232 first and then finalize this PR. It will be quite tidy to have no formal dependancies.
@jcmgray I think this is getting much closer. I updated the outstanding items in the OP.
@jcmgray and @janeyx99 This is ready for review, but likely has a few bugs to still iron out. In particular @janeyx99 is there any tests you would recommend for PyTorch...
@janeyx99 I do have a torch-only env explicitly for testing non-numpy evaluations. I like adding the check to ensure that NumPy is not present.
@rlouf Absolutely, always happy to review PRs!
My initial guess is the `opt_einsum` is deciding to use BLAS rather than einsum for this contraction. This choice is only marginally better given the size of your tensor is...
I need to confirm, but this appears to be a bug.
We should call [np.asanyarray](https://numpy.org/doc/stable/reference/generated/numpy.asanyarray.html#numpy.asanyarray) near the end of the computation which we currently do not do officially booting this to bug status. Do recall this being dropped @jcmgray ?
That is effectively the operation that something like Psi4 uses so we have very similar use cases. Let see the operation you want to complete is `C_ui G_up -> E_ip`...