Better performance than TQ and better quality than FP16?

Am I reading this right??

It's not better quality: 59.3% vs 59.4% fp16 on AIME 25

0.1% is within margin of error. Depending on the performance boost, it might be worthwhile taking a minuscule quality hit.

any divergence (even if the benchmark is better) from full precision is error

Why this is not a PR for vLLM ?

Last I heard, vLLM was backed by a company that has raised $150m in seed funding. I'm sure they've got the resources to port it.

It's the output of a research paper; the authors are not trying to build up vLLM, and they probably have no incentive to do so. You can submit a PR, though! It's easier now while the divergence is low, so don't wait. Since there are six authors, I bet you could get help with the inevitable review chores if you just take the step of creating the PR.

edit: It might not be clear that it is based on vLLM 0.22, which is the current version: https://github.com/huawei-csl/KVarN/commit/d6290e99098d7426d.... All you have to do is create a diff off it; it's fairly straightforward.

And with the help of AI, pointing at AI at this paper and saying "making a vLLM PR from this paper" tends to work surprisingly well, even if you need to nudge it a little bit along the way.

Why this is not a PR for llama.cpp

it should be easy to do btw

yao yao ling xian