brrrrrm

5 ms·

31.
Show HN: GT: experimental multiplexed distributed tensor framework(github.com)
4 points by brrrrrm 11mo ago|0 comments
32.
GT – Experimental multiplexing tensor framework for distributed GPU computing(github.com)
30 points by brrrrrm 11mo ago|1 comments
33.
by brrrrrm 11mo ago
one thing I've learned in my career is that escape hatches are one of the most important things in tools made for building other stuff. dropping down into the familiar or the simple or the dumb is so innately necessary in the building
34.
by brrrrrm 11mo ago
> None of which are they currently capable what makes you say this? modern LLMs (the top players in this leaderboard) are typically equipped with the ability to execute arbitrary Python and regularly do math + random generations. I agree
35.
by brrrrrm 1y ago
what about "test time scaling"?
36.
MFU Is Poorly Approximating Billions of Dollars in Compute(jott.live)
4 points by brrrrrm 1y ago|0 comments
37.
Model Flop Utilization Beyond 6ND(jott.live)
17 points by brrrrrm 1y ago|0 comments
38.
SWE-bench verified agents may look at future repository state(github.com)
4 points by brrrrrm 1y ago|0 comments
39.
by brrrrrm 1y ago
downshifting? these are all electric vehicles IIUC
40.
by brrrrrm 1y ago
sure, numerous examples can be shown to say smart play does help. but, would you argue the net benefits of smart play are identical between a sport like basketball and racing?
41.
by brrrrrm 1y ago
I wouldn't deny that (and probably should have caveated this in my OP), but compared to a basketball or football team, the benefit of smart play doesn't seem as significant compared to doping up and pressing hard.
42.
by brrrrrm 1y ago
> endurance sport which is a natural target for doping This makes a lot of sense to me. A very singular goal of "maximum output" without much need for fine motor skills and strategizing. I'd guess sprinting/marathon
43.
by brrrrrm 1y ago
emergent behavior. These things are surprisingly good at generalizing
44.
by brrrrrm 1y ago
what's going to be interesting is to see the large space of fused kernels being tackled by AI generated code. that might include gemm + relu + gemm + a norm of some kind - which would be annoyingly exhaustive to 1. sweep with a tuner
45.
by brrrrrm 1y ago
Open-source use of closed source models?
46.
by brrrrrm 1y ago
wait sorry, can you explain how this works? I thought gemma3 used siglip, which can output all 256 embeddings in parallel (also, would you mind sharing a code pointer if you have any handy? I found this https://github.com/
47.
by brrrrrm 1y ago
hmm, I'm getting the same results - but I see on M1 with a 7b model we should expect ~10x faster prompt processing https://github.com/ggml-org/llama.cpp/discussions/4167 I wonder if it's the encoder
48.
by brrrrrm 1y ago
> When using least squares, a zero derivative always marks a minimum. But that's not true in general ... To tell the difference between a minimum and a maximum, you'd need to look at the second derivative. It's interesting
49.
by brrrrrm 2y ago
how is this data collected?
50.
Animators Say, 'That's All, Folks' (2000)(nytimes.com)
2 points by brrrrrm 2y ago|0 comments
51.
by brrrrrm 2y ago
a premise*
52.
by brrrrrm 2y ago
Do super computers run in fp64 mostly? At fp8 an h100 hits 2 petaflops, and with only 1000 of them you’ve got more compute power than el capitan (in raw flop count)
53.
by brrrrrm 2y ago
Paper https://arxiv.org/abs/2501.00663
54.
by brrrrrm 2y ago
I probably came in too hot on that (dealing with some personal stuff). Although I disagree with purported the impact of the paper, I don’t think this is fundamentally incorrect or bad science and I wish you the best on future research.
55.
by brrrrrm 2y ago
this paper plays into some popular fantasy about the aesthetic of ANNs. it’s not scientifically useful
56.
by brrrrrm 2y ago
"y'all" a particularly popular term in NYC (I use it frequently and have lived in NYC most of my life) - so it's kinda bizarre to see "texas monthly" responding to "new york times" as if there is any
57.
by brrrrrm 2y ago
probably my biggest pure JS app is this one: https://github.com/bwasti/mebm you can try it here https://bwasti.github.io/mebm/
58.
by brrrrrm 2y ago
your single word firehose demo is awesome. I can see a lot of creative people doing really cool things with this
59.
by brrrrrm 2y ago
how are you running CUDA on the integrated Apple silicon GPU these days?
60.
by brrrrrm 2y ago
would be fun to do a leaderboard of some specific size (e.g. 4096x4096x4096) just to get all the code and tricks in one spot for folks to learn about things

More ›