roberta-large support

[alexbenari]

Is it possible to use roberta-large with llama.cpp?
My end goal is to compute a log-likelihood sentence score for sequences using this model (as defined in this paper). So I was planning to use llama.cpp for inference and implement the score computation on top of that (basically port the minicons python implementaiton or find an existing cpp implementaiton which I failed to do so far).

Does anyone have an idea how I can get roberta-large to work on llama.cpp? Any pointers on scoring implementation will also be gratly appreciated,