@amilsted can you test this out on cases where the previous code was failing?

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 80.91%. Comparing base (93dc9bd) to head (9eebb9e).
Report is 1 commits behind head on main.

@@            Coverage Diff             @@
##             main    #1646      +/-   ##
==========================================
- Coverage   80.93%   80.91%   -0.02%     
==========================================
  Files          59       59              
  Lines        4636     4626      -10     
==========================================
- Hits         3752     3743       -9     
+ Misses        884      883       -1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

We monkey-patched those changes into an older version and it indeed fixed the assertion errors we were seeing.

FWIW, just using the default_rng() also worked. Is it functionally any different than this?

We monkey-patched those changes into an older version and it indeed fixed the assertion errors we were seeing.

Great, thanks for checking.

FWIW, just using the default_rng() also worked. Is it functionally any different than this?

I want to use a separate RNG stream from the one used by the rand(elt, dims)/randn(elt, dims) array constructors, since it could cause Index ID clashes if someone sets the seed of default_rng() with Random.seed!(seed) (that's particularly an issue with saving and loading ITensors). It isn't clear to me that I can "copy" the default_rng() for that purpose, but maybe I'm missing something.

Btw, credit to @wildart for finding the issue and testing the changes.