* TensorApplier.IApplier takes ActionBuffers instead of float[] as input argument * Model output format changed