buildbot

39 ms·

601.
by buildbot 2y ago
Yeah I’d guess inference, training relies much more on high speed interconnect between nodes, which I’m sure they could do but it’s certainly another step up in complexity.
602.
by buildbot 2y ago
Yeah diffusion models are typically for image gen
603.
by buildbot 2y ago
I think the most important and somewhat massive news is that Apple built an entire LLM private cloud using Apple Silicon - that's a big deal/investment.
604.
by buildbot 2y ago
Wow - This seems at first read to be really impressive work. They got scaling laws up to a reasonable size, 2.7B, and also run a few downstream tasks. Would be interesting to see how a comparable model trained by someone else does, to check
605.
by buildbot 2y ago
Fully agree with this. Understand the basic tools that currently exist and you'll be fine. Conda constantly fucks things up in weird hard to debug ways...
606.
by buildbot 2y ago
I do all the time, and always have (in fact my first job was bare metal OS install automation), this was Rocky 9.4. New codebase, new compiler weird errors. I did actually reinstall and switch over to Ubuntu 24.04 after that issue lol.
607.
by buildbot 2y ago
A lot do, personally, every single time I try to go back to conda/mamba whatever, I get some extremely weird C/C++ related linking bug - just recently, I ran into an issue where the environment was _almost_ completely isolated fro
608.
by buildbot 2y ago
I agree, the ability to take screenshots is unsafe and should be removed. A virus is just a PRT SCRN away from stealing everything! (/s)
609.
by buildbot 2y ago
Thanks
610.
by buildbot 2y ago
In general (for those unaware) this is called stenography. You can hide an image in the lower bits of another image for example too.
611.
by buildbot 2y ago
How can you actually open and view a warc file? I've never found a good application for this, have I missed something obvious?
612.
by buildbot 2y ago
I’ve been in a small group at an academic institution. With our meager resources we trained larger models than this on many different vision problems. I personally train LLMs on OpenWebText than this using a few 4090s (not work related). Is
613.
by buildbot 2y ago
Why only MNIST and a Graph CNN? Those are small and somewhat odd choices. Scale these days should be at least 100 million param models and something like OpenWebText as a dataset in my opinion. Not sure what the SoTA is for visionm but same
614.
by buildbot 2y ago
AVX512 in a single cycle vs 2 cycles is big if the clock speed can be maintained at all near 5GHz. Also doubling of L1 cache bandwidth is interesting! Possibly, needed to actually feed an AVX512 rich instruction stream I guess.
615.
by buildbot 2y ago
An interesting but not entirely practical offshoot of this idea is a fund that buys (collects legally?) wallets with lots of crypto & lost passwords and tries to crack them over time.
616.
by buildbot 2y ago
I’m struck between the similarity between this and the big antenna (?) array that the ballon that went over the USA carried…
617.
by buildbot 2y ago
No idea if they did or not, but one immediate issue is that if anything does go wrong, now you have a nuclear incident as well as a tragedy.
618.
by buildbot 2y ago
Exactly, you have an encrypted drive? Maybe you forgot the password? Well we will just assume it’s literally the worst possible content and go from there!
619.
by buildbot 2y ago
At first read it does seem suspicious that they used Signal, however at the very end the reason is explained - it's due to the Jeff Bezos hacks in 2018. That seems very reasonable. I fully support what the FTC is doing here, however I
620.
by buildbot 2y ago
It’s really quite amazing to login a linux shell on an orangecrab FPGA running a RISV-V softcore, built using an open source toolchain. That was impossible not so long ago! At best you’d have something like Xilinx PetaLinux and all their pr
621.
by buildbot 2y ago
Counterpoint to the article - I’ve actually been slightly confused before what adobe app was purple then realized it was delta. It’s very very close. If you are colorblind, maybe even closer.
622.
by buildbot 2y ago
Yeah I don’t love the idea of an error being able to prompt inject i to a javascript console, even if it is very sandboxed…
623.
by buildbot 2y ago
Yeah, exactly. Though the 48k param lora might be as good as a 48k param layer of higher rank, I haven't looked into that case really.
624.
by buildbot 2y ago
A desolation called peace (A great phrase and book: https://en.wikipedia.org/wiki/A_Desolation_Called_Peace )
625.
by buildbot 2y ago
I’ve tested specifically this (on my personal time) :) It will train but I found the loss is proportional to the number of trainable parameters. So roughly to hit the performance of a standard 70m param model, you need to train ~70m lora pa
626.
by buildbot 2y ago
Yes, I’ve tested this out. It does train, but the scaling doesn’t seem to pan out. It’ll perform slightly better than the number of trainable parameters, but never improves as you scale, so for now there’s no benefit.
627.
by buildbot 2y ago
Slave labor and ramps. Just kidding, you and I both know that they are landing pads for Goa'uld starships so they probably used their tech. (Plus slaves).
628.
by buildbot 2y ago
Cool, instant cuda acceleration via cupy! `import cupy as np`
629.
by buildbot 2y ago
I figured ;)
630.
by buildbot 2y ago
I wasn't aware lockdown mode can be enabled without you explicitly doing so.

More ›