How can I get `tfrecord` data of proteins?

Thank you for your work to help train/fine-tune AF2/OpenFold/pLMFold models.

I tried to run pLMFold's training using my own protein datasets, but couldn't figure out how proteins' tfrecord data can be obtained.

Reading [Paper](https://academic.oup.com/bioinformatics/article/39/1/btac773/6887136), [Supplementary Data](https://oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/bioinformatics/39/1/10.1093_bioinformatics_btac773/1/btac773_supplementary_data.pdf?Expires=1699511732&Signature=jvnksD-Aa6vlcJXcb-K15QSkKTSmrgijzdMBRIlE9qD40vOW3gSd8hgNhbe3lYK--Nnni~UGlcWUIJO3PsF7hw6qR5f857AD74Rep-xJtGyz2-3ANcfigTPGVCJsoZW24tvyGB8~6HqYiXWsgOyF3lE0gccSvAiKGc9k-0A4VQ5-NTN12hwu10aGY9QuhYbOuZbOaRfluBG6fHRZaGoxJFUKGqNH9JQKVHAQLLESl-QgjRvVPh6TDb5H-5Ta0SA8DsBMwp9fVB8Pw9Jmk8m~QGl0SfH1fSvZtfZVMuNrkRfmDMm0grVsCCNpGQZUY5XZKzVl4pmZpaWzGXEnZZqYMw__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA) and [README](https://github.com/instadeepai/manyfold/blob/master/datasets/README.md) didn't help me because it has no descriptions in detail about obtaining `tfrecord` data.

I tried to make use of AF2 modules to get those data.
It seems to work but I found that some features written in the paper are missing in features generated by correspondent AF2 codes(`template_all_atom_exists` and `pdb_cluster_size`).

How could I obtain necessary features from my own proteins' data to train/fine-tune the model?
Is there any tool to do so?

I need your help.

Ref. #7