Initial model for RLHF.

Hi,

Llama 2 paper: "We initialize our reward models from pretrained chat model checkpoints, as it ensures that both models benefit from knowledge acquired in pretraining."
The "pretrained chat model checkpoints" here refers to the original pretrained model or the model after SFT. I think it should be the model after SFT. Could anyone corfirm that?

Thanks!