Regarding the limitations on the time and token count for generating an answer, I found two different answers on website (AIcrowd | Meta Comprehensive RAG Benchmark: KDD Cup 2024 | Challenges): one states 5 seconds for 50 tokens, and the other states 10 seconds for 75 bpe tokens. Which one should I consider as the standard?
Additionally, what does ‘75 bpe tokens’ mean, and does it carry the same meaning as ‘75 tokens’?
Thanks!
To your first question, I have the same confusion, you can see it at here
To your second question, yes, bpe token == token in the context of llama tokenizer