4 comments

  • throwa356262 9 hours ago
    Better performance than TQ and better quality than FP16?

    Am I reading this right??

    • qeternity 8 hours ago
      It's not better quality: 59.3% vs 59.4% fp16 on AIME 25
      • sheepscreek 21 minutes ago
        0.1% is within margin of error. Depending on the performance boost, it might be worthwhile taking a minuscule quality hit.
    • electroglyph 3 hours ago
      any divergence (even if the benchmark is better) from full precision is error
    • thefox96 8 hours ago
      Faster than Fp16, not better quality i guess
    • pbich 8 hours ago
      [dead]
  • v3ss0n 9 hours ago
    Why this is not a PR for vLLM ?
    • woadwarrior01 4 hours ago
      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.
    • electronsoup 1 hour ago
      Why this is not a PR for llama.cpp
    • esafak 9 hours 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.

      • jmalicki 8 hours 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.
    • thefox96 7 hours ago
      it should be easy to do btw
  • 0xjeffro 3 hours ago
    yao yao ling xian
  • shockembopper 7 hours ago
    [dead]