Created by: emailweixu
For trainer_count=1, the gradient machine is NeuralNetwork, which does not create parameter buf for PARAMETER_GRADIENT for sparse update in Parameter::enableType(). But gradient parameter buf is still used in SgdThreadUpdater.
Created by: emailweixu
For trainer_count=1, the gradient machine is NeuralNetwork, which does not create parameter buf for PARAMETER_GRADIENT for sparse update in Parameter::enableType(). But gradient parameter buf is still used in SgdThreadUpdater.