The softmax activation in the model I received was in three dimensions, so I added support for that layer shape to the CPU version of the softmax activation by simply extending another dimension to the library implementation. Wouldn't mind if someone takes a look at a 3D-GPU version? :1234: :)
The softmax activation in the model I received was in three dimensions, so I added support for that layer shape to the CPU version of the softmax activation by simply extending another dimension to the library implementation. Wouldn't mind if someone takes a look at a 3D-GPU version? :1234: :)