RAFT Dataset

[hieudx149]

Hello,

Your Paper RAFT is truly fantastic. It addresses the issues we've encountered when deploying generative AI applications for enterprise data. I have two questions regarding the data used in RAFT:

1. How many data samples did you use to train LLama2?
2. Could you provide me with some examples of real data you used to train LLama2? Specifically:
   • Example of data usage: P % of data: Q + D∗ + D2 + . . . + Dk → A∗
   • Example of data usage: (1 − P) % of data: Q + D1 + D2 + . . . + Dk → A∗ (when there are no 'oracle' documents, what would A* look like? I can't imagine if it's an answer like "no information to answer")

Answering these questions would greatly assist me in training our LLM. Thank you very much.