softmax_op

→TensorImplPtr

int softmax_op(const int & a, int axis)

source edit

Python wrappers

Numerically stable softmax $p_i = e^{x_i} / \sum_j e^{x_j}$ along axis.

See Also

log_softmax_op : Numerically advantageous log-domain variant for downstream NLL / cross-entropy losses. SoftmaxBackward : Autograd node implementing the gradient rule.

Parameters

aTensorImplPtr

Input logits.

axisint

Axis along which to normalise. Negative values count from the end.

Returns

TensorImplPtr

Output probabilities with the same shape and dtype as a.

Loading…

Lucid3.0