4 ms·
And may be even more adapted to sorts of RL finetuning?
by barrenko 11mo ago
And may be even more adapted to sorts of RL finetuning?
- mike_hearn 11mo agoThey say this technique isn't compatible yet with RL because you can't adjust the logits. So no GRPO I guess, which is going to be the biggest issue. An LLM with no RL applied isn't going to be that useful.