Feature Request: New sampling method that boosts reasoning performance - looks too good?

[mirek190]

Prerequisites

• I am running the latest code. Mention the version if possible as well.
• I carefully followed the README.md.
• I searched using keywords relevant to my issue to make sure that I am creating a new issue that is not already open (or closed).
• I reviewed the Discussions, and have a new and useful enhancement to share.

Feature Description

https://arxiv.org/pdf/2503.13288

Motivation

Maybe is worth to implement...

Possible Implementation

better performance

[arch-btw]

Repo: https://github.com/xufangzhi/phi-Decoding

[BrickBee]

Beam search was removed in #7736 / 0cd6bd3. This could be a nice occasion for re-adding in this more efficient form, with proper integration this time.

[mirek190]

I also think it is worth to implement.
Reading a paper seems it is not too difficult to do that.

[github-actions]

This issue was closed because it has been inactive for 14 days since being marked as stale.