mathisfun123

40 ms·

661.
by mathisfun123 3y ago
I'm sorry but I think you don't know what you're looking at - of the NN things in the top 50 there I see 4 such wrappers. The remainder are the NN frameworks and various perf motivated implementations. Also I don't know
662.
by mathisfun123 3y ago
"my wife is pregnant but my kid doesn't exist yet" hmmm
663.
by mathisfun123 3y ago
So weird - you really think AI/ML is a flash in the pan when it comes to Python? You realize pytorch, tensorflow, pandas, numpy, scikit-learn are all approaching 10 years on pypi?
664.
by mathisfun123 4y ago
Got a pointer? Paper or code either works.
665.
by mathisfun123 4y ago
Nope AITemplate is just a tracing mechanism plus cutlass plus autotuning. You still need a runtime. AIT also works on very few models (it's a replacement for something called "static runtime"). It's surprising to me how
666.
by mathisfun123 4y ago
Gonna respond here and correct both comments >Some context for those who aren't in the loop: ONNX Runtime ( https://onnxruntime.ai/ ) is a standardization format for AI models. It's just an IR, one of many - ever
667.
by mathisfun123 4y ago
>You, my friend, are dismissive to the fact that people are only willing to put in their valuable time into something if it seems at least on first glimpse interesting enough. simple solution: don't pontificate/comment if you d
668.
by mathisfun123 4y ago
>Parallel Iteration: https://en.cppreference.com/w/cpp/algorithm#Execution_polici ... You're still having trouble reading so let's try socratic method: if I have a vector with 10 elements and a vector
669.
by mathisfun123 4y ago
>Whatever is going on underneath doesn't matter and coming up with a different term is completely unnecessary. my dismissive friend: 1. who do you think implements those stl iterators whose elegance you so greatly admire? we do. 2.
670.
by mathisfun123 4y ago
> Would you mind to explain shortly what co-iteration is? it's funny how allergic hn is to reading . it's literally completely/succinctly/adequately illustrated in the Figure 1; you have a sparse array that's re
671.
by mathisfun123 4y ago
> I'm guessing they just invented the word coiteration, because if I google it reads like "i don't see any bears so the bear patrol must be working" since when i google i find plenty: https://scholar.google
672.
by mathisfun123 4y ago
this is absolutely the opposite how the citation game works - very often the first person to hit the buzzer gets all the credit and the people that come after and actually give a clean, rigorous, expository treatment gets absolutely zero cr
673.
by mathisfun123 4y ago
>I think this is more in the lines of mixing up concepts (the language features that enable implementing functors etc. in Haskell with functors as used in programming). You know how you can identify an fp person? They'll condescend
674.
by mathisfun123 4y ago
Yes monads in Haskell aren't "lawful" but they're still useful, which you rightly point out does indirectly reaffirm my claim.
675.
by mathisfun123 4y ago
https://github.com/openai/triton/pull/1358#issue-1628393794 >One fun thing - after tracking down the code to the block of C++ code, ChatGPT-4 is what actually found the memory leak for me :) although I can
676.
by mathisfun123 4y ago
As someone that knows enough cat theory and fp to scoff at most formal presentations (including all of the famous hypebeast ones hn loves to recommend), I have to say bravo to that guy (person) for doing something I've thought of doing
677.
by mathisfun123 4y ago
people hyperfocus on parsing because it's far and away the easiest part of a compiler - type inference, optimization, target codegen are infinitely harder and deeper wells to plumb.
678.
by mathisfun123 4y ago
>Hey I am the first author of one of the "abstractions" I know who you are and you should probably be out in the open with the fact that you have a conflict of interest in working at octo, a company that sells a very specific t
679.
by mathisfun123 4y ago
yup exactly; it's like other comments on hn about nn frameworks: "abstraction is the most important thing - look at pytorch it's the best framework because of the perfect/beautiful/brilliant abstractions " (re
680.
by mathisfun123 4y ago
> A properly designed compiler can allow developers to optimize matmuls within 100 lines of code. man this is such a funny closing comment - what exactly do you think is involved in designing a compiler that enables devs to optimize matm
681.
by mathisfun123 4y ago
Why did you recommend SSD at all if it doesn't make a difference?
682.
by mathisfun123 4y ago
> If you're not paging out of ram you will never ever notice. What you say is true but you didn't respond to someone complaining about ram latencies, you responded to someone lamenting poor disk access perf. And you recommended
683.
by mathisfun123 4y ago
You literally lost perf by moving to ssd through USB https://alexellisuk.medium.com/upgrade-your-raspberry-pi-4-w... >Interestingly, the SD card gave a seemingly better buffered disk read than the M2 SATA SSD at 43.35 MB
684.
by mathisfun123 4y ago
> as of 2020 (PEP 638) >> Reference Implementation >> None as yet. > but they're still slightly janky maybe you're from the future?
685.
by mathisfun123 4y ago
basically all correct but >You can then hypothetically lower (compiler terminology) it to a TensorRT MLIR dialect that then in turn runs on the Nvidia GPU. there's no tensorrt dialect (there are nvgpu and nvvm dialects) nor would th
686.
by mathisfun123 4y ago
openxla is a ML-ish compiler ecosystem built primarily around mlir that can target (through nvptx backend in llvm) and run on nvidia devices (on iree). tensorrt is a runtime for cuda programs. certainly they have features in common as a ref