Confusion about using other LLMs

Meta Comprehensive RAG Benchmark: KDD Cup 2
#1

Based on my understanding of the competition rules:
The generative model in RAG needs to be based on the LLaMA model.
During the training phase, we are able to utilize public datasets for training, which could be produced by other LLMs. However, if we opt to use additional generated data, we must generate this data using the LLaMA model, and disclose it upon completion.
I wonder if I understand the rules correctly.

#2

You may use other LLMs to generate data for training if the process follows our challenge rules and the ToS/license of the other LLMs, as answered here.

Best Regards,
The CRAG Team