Use with LLAMA

[ezzini]

Can PICARD be used with LLAMA-like models?

[nzw0301]

I suppose there are two blockers: the docker container's Python is too old, specifically, python 3.7, to use hugging face supporting llama v2 that needs python>=3.8 and the current code base assumes T5 architecture (seq2seq model) and its tokenizer.