Tokens per sample upper limit for GPTJ

[arjunsuresh]

Is there any reason why we have an accuracy upper limit for LLAMA2 Tokens per sample but not for GPT-J? It's good to document this reason for users.

[Oseltamivir]

Apparently mixtral too: https://github.com/mlcommons/inference/blob/master/tools/submission/submission_checker.py#L381

Just my two cents but might it be because these models tend to keep repeating: issue on TRT for llama, and for mixtral or maybe I have just confirmation bias.