chillee

12 ms·

91.
by chillee 5y ago
With this kind of stuff, I think the devil is in the details, but the principle is similar, yes. Specifically, the level of abstraction at which you're tracing, as well as what types of programs you can express/not express. For ex
92.
by chillee 5y ago
haha, yes, but that requires you to modify existing code to do so (which isn't always possible!). There might also be other things you want to do (like add profiling after each op) that would be tedious to do manually, but can easily a
93.
by chillee 5y ago
To give another take, Python's dynamic nature allows you to easily change class properties, methods, and so on. However, what is difficult with Python is to actually modify the executed code. For example, say you had def f(x):
94.
by chillee 5y ago
IMO, FX is more of a toolkit for writing transforms over your FX modules than "moving in Jax's direction" (although there are certainly some similarities!) It's not totally clear what "Jax's direction" mea
95.
by chillee 5y ago
Not quite - it's more of a platform for users to write their own transformations on their PyTorch code. This can include things like doing operator fusion/lowering to a backend compiler, but can also include things like insertin
96.
by chillee 5y ago
> Facebook gives you the illusion that you can opt of this data being public. What makes you think that that isn't the case here? It sounds like they just scraped all the public profiles they could find. It's not a database &qu
97.
by chillee 5y ago
https://lkyspp.nus.edu.sg/gia/article/what-did-china's-one-c... > When the policy was implemented, 26 [out of 31] Chinese provinces allowed ethnic minorities to have an additional child. There were other e
98.
by chillee 6y ago
https://news.ycombinator.com/item?id=26643703
99.
by chillee 6y ago
It's more about hardware - these models were trained on TPUs, while GPT-NeoX is being trained on GPUs graciously provided by Coreweave.
100.
by chillee 6y ago
Haha, since he didn't mention it, this guy (complexplane), is also the WR holder in super monkey ball 2: https://www.speedrun.com/supermonkeyball2
101.
by chillee 6y ago
FWIW, I really liked the sections of this book that I've read. In particular, I read the sections on "Minimum Spanning Trees" and "Fast Fourier Transform" and found them the clearest treatment of the subject that I&
102.
by chillee 6y ago
In the article, he was mentioning one specific outsider (Zvi Mowshowitz) that he trusts. I changed it to "[an outsider]" more generally since it didn't make sense without context.
103.
by chillee 6y ago
From https://astralcodexten.substack.com/p/webmd-and-the-tragedy-... > When [an outsider] asserts an opinion, he has only one thing he's optimizing for - being right. > When the Director of the CDC asserts a
104.
by chillee 6y ago
I wrote this blog post after I saw this question on HN, and somebody encouraged me to write a blog post :) https://news.ycombinator.com/item?id=22953121 Hopefully other people find this interesting as well.
105.
How Many Paths of Length K?: and I Show You How Deep the Rabbit Hole Goes(horace.io)
2 points by chillee 6y ago|1 comments
106.
by chillee 6y ago
> The moral of the story is: if you’re against witch-hunts, and you promise to found your own little utopian community where witch-hunts will never happen, your new society will end up consisting of approximately three principled civil l
107.
by chillee 6y ago
The primary issue is that large scale GPU training is primarily dominated by communication costs. Since to some approximation things need to be synchronized after every gradient update, it becomes very quickly quite infeasible to increase t
108.
by chillee 6y ago
Haha Connor (although one of the main participants) definitely isn't a BDFL - we don't have any BDFLs :) We don't really have much of a hierarchy at all - it's mostly just a collection of researchers of widely varying ba
109.
by chillee 6y ago
This is a nice tutorial: https://www.cs.cornell.edu/~asampson/blog/llvm.html
110.
by chillee 6y ago
I haven't heard anything about Google using Rust, other than for (essentially toy) usage in Fuchsia.
111.
by chillee 6y ago
https://github.com/google-research/dex-lang Another interesting language in this vein is Dex. The authors are creators of Jax and Pytorch, and they have a lot of interesting ideas.
112.
by chillee 6y ago
One issue with "getting paid for open source" is how much marginal value the project is generating. For example, take left-pad. At the time of its infamous removal, it was relied upon by thousands of packages, including Node and B
113.
Combining Label Propagation and Simple Models Out-Performs Graph Neural Networks(arxiv.org)
1 points by chillee 6y ago|1 comments
114.
by chillee 6y ago
Also see this tweet thread summarizing the paper: https://twitter.com/cHHillee/status/1323323061370724352 , as well as a higher level of where this paper fits into the general literature here: https://tw
115.
by chillee 6y ago
That's not GPT-3.
116.
by chillee 6y ago
In this case, I think the 2 are aligned.
117.
by chillee 6y ago
Reminds me of that Grouchism: > Nobody goes there anymore, it's too crowded
118.
by chillee 6y ago
One issue now is that the title mentions Facebook but the linked article does not, which is somewhat confusing.
119.
by chillee 6y ago
That doesn't match my experience. I looked at the first and 5th page from my link. No spam PRs. https://i.imgur.com/VgOteKy.png https://i.imgur.com/aAZ8xbo.jpg
120.
by chillee 6y ago
I'm trying to get a lower bound on spam PRs filed. If you sort by oldest (instead of by "relevance" as the parent commenter did) and do a spot check, you'll see that a lot of the PR's are legitimate: https:/&#

More ›