thanks so much @psobolewskiPhD !! the MPS error looks strange to me -- it says it's out of memory? otherwise the windows errors are expected due to changes in the dynamics code resulting in different numbers of cells from cellpose v2 and beyond

Yea sorry I just left it last night, I was tracking down the failures and got sleepy.
The actual test errors are indeed just because the defaults have changed. I will explicitly set the model to prevent that in the future.
The MPS fail is this:
actions/runner-images#9918
which points to the runner docs: https://docs.github.com/en/actions/using-github-hosted-runners/using-github-hosted-runners/about-github-hosted-runners#limitations-for-arm64-macos-runners
It's interesting that cellpose proper tests don't hit this? I will take a look, maybe y'all have a clever solution there.
Otherwise I think the solution is to monkey patch MPS availability on CI.

Thanks! Ah no we did also have failing regression tests, and I've updated the expected output with each release, so we can do that here too by re-setting the assert values - or I can change it to the same regression tests as cellpose (although I don't have a diameter test there)

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 60.08%. Comparing base (ef6d108) to head (dacf1e4).
Report is 1 commits behind head on main.

@@            Coverage Diff             @@
##             main      #55      +/-   ##
==========================================
+ Coverage   54.31%   60.08%   +5.77%     
==========================================
  Files           3        3              
  Lines         232      233       +1     
==========================================
+ Hits          126      140      +14     
+ Misses        106       93      -13     

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

Ok, things are working.
The logic here: https://github.com/MouseLand/cellpose/blob/52f75f9636250979dcff48da18d24aaff41c2c86/cellpose/core.py#L90-L108
Doesn't really handle passing gpu=True and yet having mps backend is_available() return False because of the nested if in try (which doesn't except).
So I monkeypatched that function too to be safe.

The diameter value I get when running locally with MPS is different than with CPU, so I split the difference so that I can run tests locally. And I explicitly set cyto3 and the number of cells I get with that.

I'll try re-enabling Ubuntu tests next.

@carsen-stringer I think it's ready for review now 😃

thank you so much! 🔥 🚀