Alternative of CachedDataset in torch_xla for PyTorch?

Is there any implementation of analogy of CachedDataset in pytorch library? Or on github?