Contrastive decoding improves reasoning in large language models

[mfiguiere]

https://arxiv.org/abs/2309.09117

[Llamamoe]

I wonder how big of an improvement could be achieved by combining all the different approaches like this together.