Hi, Thanks for sharing your code.
I wonder where the iterative self-improvement is happening in your code.
Are SFT and RL performed offline, or are they run iteratively? If it’s the latter, could you point me to the relevant code repository?
Thanks