HackerTrans
TopNewTrendsCommentsPastAskShowJobs

lukebechtel

no profile record

Submissions

Surpassing vLLM with a Generated Inference Stack

infinity.inc
62 points·by lukebechtel·4 maanden geleden·22 comments

Aristotle's Rhetoric, Strunkified

lukebechtel.com
4 points·by lukebechtel·6 maanden geleden·3 comments

Zeroed Out

lukebechtel.substack.com
7 points·by lukebechtel·7 maanden geleden·1 comments

Broken Koans

davidchess.com
1 points·by lukebechtel·9 maanden geleden·0 comments

Vasty Houses (2001)

davidchess.com
1 points·by lukebechtel·9 maanden geleden·0 comments

Environmental Epigenetics

en.wikipedia.org
5 points·by lukebechtel·10 maanden geleden·0 comments

comments

lukebechtel
·2 maanden geleden·discuss
[dead]
lukebechtel
·3 maanden geleden·discuss
[dead]
lukebechtel
·3 maanden geleden·discuss
Some people seem to be suggesting these are coverups for quantization...

Those who work on agent harnesses for a living realize how sensitive models can be to even minor changes in the prompt.

I would not suspect quantization before I would suspect harness changes.
lukebechtel
·3 maanden geleden·discuss
[dead]
lukebechtel
·3 maanden geleden·discuss
really fun :) thanks!
lukebechtel
·3 maanden geleden·discuss
it sounds like you described it pretty well!
lukebechtel
·4 maanden geleden·discuss
~/.claude/projects is where the real fun is :)
lukebechtel
·4 maanden geleden·discuss
What is your domain?
lukebechtel
·4 maanden geleden·discuss
so we need to make some crazy llms...
lukebechtel
·4 maanden geleden·discuss
there used to be https://default-filename-tv.neocities.org/ but it got taken down :/
lukebechtel
·4 maanden geleden·discuss
The bitter lesson strikes again, I suppose!
lukebechtel
·4 maanden geleden·discuss
Good questions! It's clear I need to gather more metrics from our next generated inference library.
lukebechtel
·4 maanden geleden·discuss
Unfortunately it hasn't been open sourced. We're debating how / when to do this right now.
lukebechtel
·4 maanden geleden·discuss
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.
lukebechtel
·4 maanden geleden·discuss
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!
lukebechtel
·4 maanden geleden·discuss
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.
lukebechtel
·4 maanden geleden·discuss
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!
lukebechtel
·4 maanden geleden·discuss
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.
lukebechtel
·5 maanden geleden·discuss
I also do this!
lukebechtel
·5 maanden geleden·discuss
Thank you Georgi <3