fn
huber_loss_op
→TensorImplPtrint huber_loss_op(const int & input, const int & target, double delta, int reduction)Python wrappers
Public Python APIs implemented by this engine symbol.Public Huber (smooth L1) loss entry point.
Parameters
inputTensorImplPtrPredicted tensor.
targetTensorImplPtrGround-truth tensor of the same shape.
deltadoubleQuadratic / linear transition threshold; must be
> 0.reductionint0 (None), 1 (Mean), or 2 (Sum).Returns
TensorImplPtrReduced or per-element loss tensor.