device cpu | train (178614, 49, 9, 1) (75 MB uint8) | val (4941, 49, 9, 1) | bins 49 win 9
params: 338,302
ep  0  loss 7.7432  val cell-acc 0.7552  val note(str+fret)-acc 0.1933
ep  2  loss 2.5779  val cell-acc 0.8684  val note(str+fret)-acc 0.6916
ep  4  loss 2.1270  val cell-acc 0.8852  val note(str+fret)-acc 0.7657
ep  6  loss 1.9533  val cell-acc 0.8864  val note(str+fret)-acc 0.7722
ep  8  loss 1.8558  val cell-acc 0.8946  val note(str+fret)-acc 0.7953
ep 10  loss 1.7913  val cell-acc 0.8923  val note(str+fret)-acc 0.7773
ep 12  loss 1.7444  val cell-acc 0.8918  val note(str+fret)-acc 0.7820
ep 14  loss 1.7042  val cell-acc 0.8970  val note(str+fret)-acc 0.8084
ep 16  loss 1.6539  val cell-acc 0.8967  val note(str+fret)-acc 0.7938
ep 18  loss 1.6371  val cell-acc 0.8923  val note(str+fret)-acc 0.7851
ep 20  loss 1.6228  val cell-acc 0.8914  val note(str+fret)-acc 0.7878
ep 21  loss 1.6154  val cell-acc 0.8967  val note(str+fret)-acc 0.7867
/Users/christianstrobele/code/onnx_runtime_dart/tool/tab_labeler/train.py:209: DeprecationWarning: You are using the legacy TorchScript-based ONNX export. Starting in PyTorch 2.9, the new torch.export-based ONNX exporter has become the default. Learn more about the new export logic: https://docs.pytorch.org/docs/stable/onnx_export.html. For exporting control flow: https://pytorch.org/tutorials/beginner/onnx/export_control_flow_model_to_onnx_tutorial.html
  torch.onnx.export(
FINAL (best ckpt)  train note-acc 0.8261  val note-acc 0.8084  (best val 0.8084)
exported /private/tmp/claude-501/-Users-christianstrobele-code-onnx-runtime-dart/9356a989-c7fb-455b-b6f8-bfa444e4d5b3/scratchpad/tab_labeler/rerun-20260721-105128/tab-labeler.onnx
wrote parity fixture (240 examples)