[Der_Einzige]

Min_p author here: I’m convinced that the whole field critically misunderstands temperature (I.e temperatures limited to 2 is very harmful for diverse generation). Articles like this are excellent and very cool.

Hacking your LLM inference engine to enable cool sampling tricks is the definition of AI research/engineering. We need more of this and less prompt grifting.

[wolttam]

Okay, something just tweaked in my brain. Do higher temperatures essentially unlock additional paths for a model to go down when solving a particular problem? Therefore, for some particularly tricky problems, you could perform many evaluations at a high temperature in hopes that the model happens to take the correct approach in one of those evaluations.

Edit: What seems to break is how high temperature /continuously/ acts to make the model's output less stable. It seems like it could be useful to use a high temperature until it's evident the model has started a new approach, and then start sampling at a lower temperature from there.

[wongarsu]

Decaying temperature might be a good approach. Generate the first token at a high temperature (like 20), then for each next token multiply temperature by 0.9 (or some other scaling factor) until you reach your steady-state target temperature

[GRiMe2D]

I think yes. Recently I was experimenting with NEAT and HyperNEAT solutions and found this site. At the bottom it explains how novelty yields far more optimal solutions. I would assume that reasonably high temperature may also result more interesting solutions from LLM

https://blog.lunatech.com/posts/2024-02-29-the-neat-algorith...

[bjourne]

Correct me if I'm wrong, but the problem is that it is almost impossible to evaluate sampling methods. You can't just look at perplexity and conclude that A is better than B. So you need large-scale expensive human evaluations. Even if you have those it is difficult to extrapolate results since what sampling method works best depends on the dataset(s).

[programjames]

I think you can try maximizing the free energy E[reward] + temperature*entropy?

[bjourne]

How do you know that generates high quality text?

[programjames]

It generalizes better, so it ought to produce higher quality text.