mathisfun123

31 ms·

511.
by mathisfun123 3y ago
Lololol show me an "MLIR" port. Do you mean tensorflow port or jax port or torch port (that uses torch-mlir)? Or do you really mean llama implemented in linalg/tosa/tendor?
512.
by mathisfun123 3y ago
AWS obviously
513.
by mathisfun123 3y ago
You guys are talking past each other but really talking about the same thing - arithmetic intensity. You're talking about FEA or some other grid solver/discretized PDE/DFT type thing where the matmuls are small because the me
514.
by mathisfun123 3y ago
> it talks a lot about how GPUs work it's true - out of all of the "LEARN CUDA IN 24 HOURS" books, this is the best one. indeed this isn't one of those same books - this is a textbook - but at first glance it resemble
515.
by mathisfun123 3y ago
> Do you intend to give the air of smug self importance? It's tacky. there's a very weird, counterintuitive, psychological phenomenon on display here. somehow, in your opinion, the article itself doesn't cross the threshol
516.
by mathisfun123 3y ago
>Peer review does not serve to assure replication, but assure readability and comprehensibility of the paper. I have had a paper rejected twice in a row over the last year. Both times the comments include something like "paper was v
517.
by mathisfun123 3y ago
> to hear what you think about it my comment had two sentences, both ending with question marks. did you know that sentences that end in question marks aren't statements (of fact or opinion)?
518.
by mathisfun123 3y ago
The misunderstanding you have is that often the article isn't the point. Just like a football match in a pub isn't always the point. People just want to chime in - the comment section is like an ephemeral pub. If that bugs you bec
519.
by mathisfun123 3y ago
if i walk by a pile of shit in my neighborhood and make a comment on it, i'm not the same as the person that took/excreted the shit in my neighborhood.
520.
by mathisfun123 3y ago
What compels someone to write this kind of banal pontification? More importantly, what compels someone to read it and then suggest it to others?
521.
by mathisfun123 3y ago
First of all you don't need a firm grasp of ito integration to make money at a hedge fund, not even 20MM. Conversely it's not like they just give you 20MM once you've passed your quals. Second of all it's really not th
522.
by mathisfun123 3y ago
> I prefer to use as few technologies as possible. And completely stay away from propriatary technologies. This is the most boring and tedious archetype out there.
523.
by mathisfun123 3y ago
Yes https://github.com/iree-org/iree-samples/blob/main/requireme... You can do even better using GitHub releases: the GET for a releases page with assets (ie use inspector) functions as an index url. So
524.
by mathisfun123 3y ago
That tut only shows you how to remix existing operators. That obviously won't work here (not what's required). You need to do this https://pytorch.org/tutorials/advanced/dispatcher.html More complex but
525.
by mathisfun123 3y ago
Yes but as you can see by reading, this comment chain starts with someone linking to a xilinx product
526.
by mathisfun123 3y ago
> Xilinx's DSP58 blocks (Versal devices) this is correct (i am currently working on designs that target these parts) but those things are more expensive than comparably sized/equipped nvidia gpus so (in the context of this disc
527.
by mathisfun123 3y ago
> I think it's more that the compile was unconstrained, so no floorplanning or handholding synthesis anything i mean they don't say anything about how long they let it run for - constraints are to narrow the search space for hu
528.
by mathisfun123 3y ago
> The mind blowing part of all of this is the fact that they were able to close timing at 771MHz It's true but I mean this is Intel in-house research right? If they can't get absolute peak fmax on their own parts that would be
529.
by mathisfun123 3y ago
I wish people would stop quoting marketing material as some kind representation of what they know. You're giving completely the wrong impression about dsp slices - it is absolutely not 1 dsp slice per FP operator at any precision that
530.
by mathisfun123 3y ago
How many lines of mojo have you written?
531.
by mathisfun123 3y ago
Who are all these people that so strongly believe in mojo before even having written a single line of it (yes I'm aware you can write "mojo" in their walled-garden). Like they've gotta be paid shills because I can't
532.
by mathisfun123 3y ago
Have you seen https://github.com/RosettaCommons/binder ? python aside, having gone down this rabbithole, and still not infrequently revisiting said rabbithole, I don't believe using *clang like this is a winning s
533.
by mathisfun123 3y ago
i said ported not implemented; the likelihood that any of those libraries sprout lisp bindings is about as likely as them being rewritten in lisp. so it's the same thing and the point is clear: i don't care about some zany runtime
534.
by mathisfun123 3y ago
>I'm glad to hear you can hook opcodes like that, then you really can do anything Just in case someone comes around calls me a liar: the way to do this is to spread the bytecodes out one per line and set a line trace. Then when your
535.
by mathisfun123 3y ago
At no point in time did I claim to know lisp well. I stated my familiarity at the outset. But what you all did was claim to know a lot about every other interpreted runtime without a grain of salt. >Next you can find out what optimizing
536.
by mathisfun123 3y ago
Okay well when pytorch, tensorflow, pandas, Django, flask, numpy, networks, script, xgboost, matplotlib, spacy, scrapy, selenium get ported to lisp, I'll consider switching (only consider though since the are probably at least another
537.
by mathisfun123 3y ago
Do you think the magic fairies are doing it for you? Your interpreter/runtime is still doing it whether you're aware of it or not. My point is very simple: I can do it too, in any language I want, and so there's nothing speci
538.
by mathisfun123 3y ago
The other guy up above claims this is a feature unique to calling functions, rather than all error states, and that the lisp runtime specifically guards against this. If that's the case then my answer is very simple: it would be trivia
539.
by mathisfun123 3y ago
>Common Lisp programs run by default in a way that calls to undefined functions are detected. Cool so what you're telling me is that by default every single function call incurs the unavoidable overhead of indirecting through some l
540.
by mathisfun123 3y ago
It works in code compiled from c++ too: define and associate a signal handler for sigkill, call a function whose symbol can't be runtime resolved by the linker, sigkill is sent and caught, define your function (in your asm dejure), pa

More ›