Rc2 improv

Correct a bug in to_onnx (to convert intermediate models) Correct the L1/L2 switch add info on norm and LR during training

Merge request reports

Loading