Enhance Dropout
The current implementation of the Dropout operator in AIDGE does not support:
- The optional third input training_mode (introduced - in ONNX opset 12)
- The secondary output mask, which indicates which elements were retained after dropout
- The seed attribute, used to seed the random number generator for dropout to ensure reproducibility
According to the ONNX specificationfor Dropout (opset ≥ 12), both features are part of the standard operator definition.
Edited by Houssem ROUIS