FYI there is also a tfrecord reader in xla: https://github.com/pytorch/xla/blob/master/torch_xla/utils/tf_record_reader.py
Have you compared this implementation to other ones? I guess it is good enough, so I will give it a try.
I’ve tried a few then but could not get anything working reasonably with multiple files, unfortunately…
I wonder if we can actually use tf.data.TFRecordDataset and convert like
Please let us know if you find a good way.
I think it’s a common request, and I (and perhaps many more) would very interested to know!