The foundational runtime components of TensorFlow make it much slower than PyTorch when replicating interactive computation features, despite efforts to imitate it using eager.