com.intel.analytics.bigdl.optim
Optimizer cache some metadata on each executor
cached models
weights of the cached models
gradients of the cached models
cached criterion
cached state
tensor buffer