Skip to content

Draft: add half precision roundoff testing utility

Dan Guest requested to merge dguest/athena:htest into main

Attempt at getting at atlas-flavor-tagging-tools/training-dataset-dumper#168

Problem is I can't seem to reproduce the weirdness we are seeing there. Dumping a 1M x 10 array of floats together with a 1M x 10 array of halves, and they all agree within 0.001 relative difference. 🤷

Merge request reports