Hi, sorry I'm in another meeting...

Quick update: Testing MIGraphX as a new inference framework supported directly for ROCm devices. Loading an ONNX model is possible, but program fails when trying to aquire the GPU for model compilation...

Working on it... But it's very surprising that not even ROCm's own library for ML models works without problems (I'm using basic, standard code from the documentation). Works neither on epn000 (rocm 5.5.3) nor on epn146 (dedicated node with rocm 6.1.0). Also we should discuss soon what Nicodemos should do. We are both basically stuck now as no framework seems compatible...