omerhac

Special thanks to https://x.com/kohya_tech for his amazing open source software, that we couldn't have done without, and to https://x.com/modal_labs for their game changing serverless GPU calls, that enable a small startup like ours to produce cutting edge results in an affordable rate ♥

Great story. Did they fire her for that one time she didn't show up? Thats tough

I'm sad to hear that you are having a hard time navigating the market. If it's of consultation know that many developers and engineers go through this hardship, those who never did either: Never pushed their limits, are liars, are Linus Trovalds. My personal thoughts about this is that I would prefer an applicant's motivation to be more in the sense of "What technologies are emerging, and what makes them so special? I want to master them to feel that I understand how they work". It's a wide answer, but I think the right engines are really what makes the difference. I would suggest finding a field you can develop passion to, preferably an obsession. Having an obsession is a great way to be extremely competent in a filed. Goof luck!

Cool tool. I know some great guys who built a successful startup out of the observability and safe kernel sandboxing you can get with eBPF. Wonder if it could be a good basis to build an osx version of win-internals.

I admit to be on the side that is very suspicious about journalists, but I haven't given enough thought as to the role of money in their bias. Money obviously trumps everything (pawn not intended :) ), but there's also the question of where the money comes from, or what money is to be made from applying certain manipulations on the public. All in all I think the notion of an 'objective' media is absurd to the point of puritanism. In democracies, the media is just another political force, like the parliament or the government, only it is not appointed by elections and less regulated, but also has no actual force, only 'soft force' (which in some cases is stringer than 'force force').

I agree, although a group of 34 people if sampled correctly can yield a statistical significant signal. My main problem with the methodology here is that it is completely qualitative with no reason. It would have been very simple to do a quantitative analysis and reach numbers that have statistical significance.

Gonna use this when I practice drums

"We plead guilty for our poor test coverage"

Each time research about LLM and reasoning comes out Yan LeCun gets an itch

Great app. A good example why js apps can never be more than OK, and native apps can really rock. Would be amazing if I could use my sublime bindings here

All you really need is a Censys account https://support.censys.io/hc/en-us

Wow this is much needed. Thanks! I keep hearing rumors that a no GIL version is coming, I think this is probably the biggest change so far, as it would make the language much more performance oriented. I'm considering a tech stack and thinking about both python and Mojo for ML, not sure that Mojo is stable yet, but the performance differences are HUGE.

Cool blog, thanks!

I did a similar project a couple of years ago for a university course, only I also added style transfer, it turned out pretty cool. I scraped a bunch of news data together with it's news section and trained a self attention language model from scratch, turned out pretty hilarious. Data was in Hebrew, which is a challenge to tokenize because of the morphology. I posted it on ArXiV if someone's interested in the style transfer and tokenization process: https://arxiv.org/abs/2212.03019

Thats cool but not exactly the same as reading text that was written to be consumed as text. I think a big part of the reason technical text is easier to digest is the way it's written, and not so much the medium.

yep, thought exactly the same thing

Hahaha, laughed my ass off.

Can you name the largest factor to why you did not ship it? Is it fear? shyness? Thinking that you only have 1 chance with your clients?

Very cool, but why is it called diffusion forcing?

This is such a valuable piece. I've learned so much reading it! And your open-source code is great as well.

Some open questions I have: 1) Why did you choose to setup your own cluster? How was the experience with your cloud partner regarding faulty machines / switches? 2) What were your considerations choosing the cluster architecture that have proven the most valuable ? (apart from the all2all comms) 3) Can you share a bit more about your logging infra apart from the fact that it was Loki based? 4) What necessitated the use of a local docker registry? did you use other images apart from nvidia-container-runtime?

Thanks!

Very cool - which method do you use for editing the images? is it SDEdit or InstructPix2Pix? another one?

I love the minimalistic UI and keyboard shortcuts - cool!