Using <sep> token in ESM 1b

[raimondifranc]

Dear all,

first of all I'd like to thank you for your amazing work and for making it publicly available.
I wonder whether it is possible to use the token, which I see is available in the Alphabet class, to get the embeddings from the ESM-1b model for concatenated sequences, for example of interacting proteins, similarly to what has been done here:
https://dl.acm.org/doi/pdf/10.1145/3388440.3412467

Many thanks in advance.
Best wishes,
Francesco Raimondi