alt Hacker NewsI am failing to see how this stops pre-training scrapping? It still looks like legit code, playing nicely with the desired pre-training distribution. Obviously nobody is going to use it for SFT/DPO/GRPO later.
I am failing to see how this stops pre-training scrapping? It still looks like legit code, playing nicely with the desired pre-training distribution. Obviously nobody is going to use it for SFT/DPO/GRPO later.