ONE - On-device Neural Engine
|
Class to make negative gamma of batchnorm to a small positive value (1e-10) This pass can change the execution result of the model. So, use it only when the impact is known to be acceptable. More...
#include <MakeBatchNormGammaPositivePass.h>
Public Member Functions | |
const char * | name (void) const final |
bool | run (loco::Graph *g) final |
Public Member Functions inherited from logo::Pass | |
virtual | ~Pass ()=default |
Class to make negative gamma of batchnorm to a small positive value (1e-10) This pass can change the execution result of the model. So, use it only when the impact is known to be acceptable.
Definition at line 30 of file MakeBatchNormGammaPositivePass.h.
|
inlinefinalvirtual |
Reimplemented from logo::Pass.
Definition at line 32 of file MakeBatchNormGammaPositivePass.h.
|
finalvirtual |
Make negative gamma values of Mul-Add (as BatchNorm) to a small positive value (1e-10)
PATTERN: | [CircleNode] [CircleConst](as gamma) | | [CircleMul] [CircleConst] | | [CircleAdd] |
Implements logo::Pass.
Definition at line 84 of file MakeBatchNormGammaPositivePass.cpp.
References loco::active_nodes(), and loco::output_nodes().
Referenced by package.infer.session::inference().