Add Shrek (entropix) Sampler

[underwaterbepis]

https://github.com/xjdr-alt/entropix

Early testing on Twitter has shown it makes Llama 3.2 1B much better at reasoning as it gives the model “longer to think” during tokens it is uncertain about

Repo and method still in development, see Twitter user xjdr (“AI Shrek”)