Updates on ONNX runtime implementation
- Draft-PR open on alidist for the build of ORT on systems with Nvidia / ROCm GPU's. Checking currently if we can build for both at the same time (with a change to https://github.com/microsoft/onnxruntime/blob/main/cmake/CMakeLists.txt)
- Draft-PR open on O2 for ORT linbrary implementation (works, and would deliver ~25-30 mio. clusters / s)
Updates on NN side
- Put QA and NN training into suite submissions, so if any change is made to training data, submission can now be performed for any network architecture and configuration as a "suite" (don't need to submit them individually)
- Eval. on real data causes some issues with missing files (o2simdigitizerworkflow_configuration.ini, ctpdigits.root)
- Training converges well, also for N-class classification, revealing sampling bugs

Updates on simulation
- Neutron capture simulation now included, so can basically generate loopers and enhance training data to classify for exclusion