KVarN: Native vLLM backend for KV-cache quantization by Huawei
56 points
2 hours ago
| 2 comments
| github.com
| HN
throwa356262
1 hour ago
[-]
Better performance than TQ and better quality than FP16?

Am I reading this right??

reply
qeternity
41 minutes ago
[-]
It's not better quality: 59.3% vs 59.4% fp16 on AIME 25
reply
thefox96
44 minutes ago
[-]
Faster than Fp16, not better quality i guess
reply
v3ss0n
1 hour ago
[-]
Why this is not a PR for vLLM ?
reply
esafak
1 hour ago
[-]
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.

reply
jmalicki
1 hour ago
[-]
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.
reply
thefox96
17 minutes ago
[-]
it should be easy to do btw
reply