Connor T. Jerzak

Thank you for these thoughts and quick work! The `ValueError: 'EmitPythonCallback'` is now gone. I've played around with a few different things (the Metal backend seems to require float32). Still...

GetElementFromTfRecordAtIndices rebuilt; test to see if rebuild fixes this.

I will send you in a minute a script that manually re-creates the environment we can do further testing on. My best guess is that we might need to set...

What happens if you try to build a new environment using conda from the terminal? If that succeeds but the BuildBackend() call fails, the problem likely lies in the setting...

Sounds good! Was this the BuildBackend() or Python way?

Cool. If the BuildBackend way doesn't work that's an indication that we need to tweak the environmental variables R is using.

Okay, noted. Yes the tf record thing is a little weird. If it still poses trouble we can move to another format. I used two envs ultimately On Fri, Jul...

Very true; in general most of these packages will require different versions for arm64 vs. x86_64 (as well as for METAL vs. Nvidia vs. AMD GPU). Often, this is handled...

Thank you! Mostly we were just curious to see if we could further optimize the model in terms of runtime (by running the model with larger batches of images and...