drop_path_op

→TensorImplPtr

int drop_path_op(const int & a, double p, bool scale_by_keep, Generator * gen)

DropPath — per-sample stochastic depth — free-function entry point.

Thin wrapper that delegates to DropPathBackward::forward.

aTensorImplPtr

Input tensor of rank

\geq 1

; first axis is the batch axis.

pdouble

Per-sample drop probability in

[0, 1)

scale_by_keepbool

Whether to rescale surviving samples by

1 / (1 - p)

genGenerator*

Optional RNG; nullptr selects default_generator().

TensorImplPtr

Output tensor of the same shape as a.

Loading…