This PR tends to add a subcommand for the optimization of nnp model in protograph level.
Currently, this feature is only for the optimization of the output from tensorflow importer, which break tf.dense() into pieces.
This feature can collect them and fuse them to an Affine().
This PR tends to add a subcommand for the optimization of nnp model in protograph level. Currently, this feature is only for the optimization of the output from tensorflow importer, which break tf.dense() into pieces. This feature can collect them and fuse them to an Affine().