Surpassing vLLM with a Generated Inference Stack
infinity.inc

・

57 points

・

lukebechtel

・

a day ago

23 comments

ntonozzi ・ a day ago

Why do they need to run benchmarks to confirm performance? Can't they run an example prompt and verify they get the exact same output token probabilities for all prompts? The fact that they are not doing this makes me suspicious that they are in fact not doing the exact same thing as vLLM.

It is also a bit weird that they are not incorporating speculative decoding, that seems like a critical performance optimization, especially for decode heavy workloads.

  • nyrikki ・ 2 hours ago

    > Can't they run an example prompt and verify they get the exact same output token probabilities for all prompts?

    You don’t even get that with GPUs in general, or really floating point in general.

    The Art of Computer Programming. Volume 2: Seminumerical Algorithms section 4.2.2 with explain where it loses floating addition associativity property.

    Apartness relations are another possible lens.

  • lukebechtel ・ a day ago

    Yes, speculative decoding will make both us and VLLM faster, but we believe it would be a relatively even bump on both sides, so we didn't include it in this comparison. Worth another test!

  • jeeeb ・ 11 hours ago

    > It is also a bit weird that they are not incorporating speculative decoding

    Wouldn’t speculative decoding decrease overall throughput, but optimise (perceived) responsiveness?

    • YetAnotherNick ・ 11 hours ago

      For compute bound region(high batch size) yes, but for low batch size it could improve the throughput.

2001zhaozhao ・ 14 hours ago

Every example like this makes it obvious that you can now use ML-like optimization approaches on well-specified, very-well-tested software problems with a clear optimization goal. Keep if it improves the objective while maintaining correctness, discard if it doesn't. AI-descent strikes again.

Maybe I should learn more about ML to have a better instinct on optimization methods in general, so I can actually build AI optimizers like these.

  • lukebechtel ・ 3 hours ago

    The bitter lesson strikes again, I suppose!

storus ・ a day ago

Does it support paged attention like vLLM though? Without that they will run into memory fragmentation quickly.

  • lukebechtel ・ a day ago

    Yes, great question!

    The system started without paged attention, and recreated its own paged attention implementation automatically once it realized it was a bottleneck.

    Pretty cool!

rfw300 ・ a day ago

OK... we need way more information than this to validate this claim! I can run Qwen-8B at 1 billion tokens per second if you don't check the model's output quality. No information is given about the source code, correctness, batching, benchmark results, quantization, etc. etc. etc.

  • lukebechtel ・ a day ago

    We validate with MMLU and Hellaswag presently, and are getting this independently verified by a 3rd party.

    We have considered open-sourcing some of our optimized inference libraries in the future, but have not yet come to a decision on this.

    Also if you need a rough intuition as to why this is possible: it's because this entire inference stack was built for exactly one model, and thus we can really tune the entire framework accordingly.

    • rfw300 ・ 20 hours ago

      I've no problem with the intuition. But I would hope for a lot more focus in the marketing materials on proving the (statistical) correctness of the implementation. 15% better inference speed is not worth it to use a completely unknown inference engine not tested across a wide range of generation scenarios.

      • lukebechtel ・ 17 hours ago

        This is a fair critique! We plan to use our system to generate many more inference libraries of this nature, and I'll make it a point to release better, broader correctness measures when we do so.

hoerzu ・ 9 hours ago

What's the jitter what's the std? What about 1:1 output equality?

What's the post request latency of this part? What the ftt?

  • lukebechtel ・ 3 hours ago

    Good questions! It's clear I need to gather more metrics from our next generated inference library.

ismailmaj ・ 9 hours ago

Any place we can find the code?

  • lukebechtel ・ 3 hours ago

    Unfortunately it hasn't been open sourced. We're debating how / when to do this right now.

    • ismailmaj ・ 2 hours ago

      Confusing, since this is specific to an architecture that no one making money will use (8B is consumer space, not enterprise). The produced code shouldn't hold much interesting IP?

acuozzo ・ a day ago

Luke: Do you have benchmarks for BF16?

  • lukebechtel ・ a day ago

    Unfortunately, not at present; we went for FP8 because we believed it was generally the best tradeoff of quality and speed. Allowed faster iteration as well.

    We believe our improvements would hold on BF16, but let me check.