The previous model does. https://arxiv.org/abs/2309.00267 https://arxiv.org/abs/2309.00267 That's during fine-tuning though. I know I read one about applying

The previous model does. https://arxiv.org/abs/2309.00267 https://arxiv.org/abs/2309.00267 That's during fine-tuning though. I know I read one about applying it during pretraining, maybe this one? https://arxiv.org/abs/2302.08582 https://arxiv.org/abs/2302.08582