Is the TCR embeddings generation code still compatible with the Zenodo model weights?

[nodrogluap]

First, let me thank you very much for this excellent resource!

I have a question:

When using the latest git HEAD, I run tdi.tl.get_pretrained_tcr_embedding(...) with the following bert_config:

tdi.model.modeling_bert.get_human_config(bert_type='small', vocab_size=36)

and using the checkpoint file human_bert_pseudosequence.tcr_v2.ckpt, I receive the following warning:

Warning: model.bert.embeddings.position_ids not found in the model. Ignoring model.bert.embeddings.position_ids in the provided state dict.

Is this something with which I should be concerned? Will the X_tcr values generated still be comparable to those in human_tcr_reference_v2.h5ad?

Thank you!