buildbot

35 ms·

691.
by buildbot 3y ago
Oh nice, good catch!
692.
by buildbot 3y ago
Plus the README.md that is just a rickroll
693.
by buildbot 3y ago
I don't know, way outside my domain. Possibly none I guess?
694.
by buildbot 3y ago
Interesting, seems to be a tiny finnish hosting company: https://www.zoner.fi/english/
695.
by buildbot 3y ago
> tukaani.org has address 5.44.245.25 (seemingly in Finland) Hetzner?
696.
by buildbot 3y ago
If libarchive is also backdoored, would that allow specifically crafted http gzip encoded responses to do bad things?
697.
by buildbot 3y ago
Yeah I saw that - I wouldn't bet on them being in the US but who knows. Maybe they just really love CRC32 ;) And introducing backdoors (if it that was them not an account takeover).
698.
by buildbot 3y ago
XNOR-Net was the "first" in that generation, as I recall. It's not a new idea at all though. Check out the dates on these papers - https://ieeexplore.ieee.org/abstract/document/286901 < 1994 ht
699.
by buildbot 3y ago
In the article it says CISA was notified - that sounds like it's going to be a federal investigation if nothing else. If I was this person, I wouldn't be in the USA (or any US friendly nation) ASAP.
700.
by buildbot 3y ago
This potentially could be a full automated rootkit type breach right? Great - is any system with 5.6.1 possibly vulnerable? Also super weird a contributor thought they could slip this in and not have it be noticed at some point. It may poin
701.
by buildbot 3y ago
Seattle also has this problem and our sewer network is more modern. We are also building a storage tunnel!
702.
by buildbot 3y ago
They also don’t default to RAW, which uses like 60-80mb a shot with 48mp! That does use storage quite fast!
703.
by buildbot 3y ago
In theory the built in files app will work for this. However, I like goodnotes, which has good highlighting snd library support. I’ve used it since grad school for reading papers.
704.
by buildbot 3y ago
Llama.cpp and many others support 4 bit weights and lower
705.
by buildbot 3y ago
Yes this is it!
706.
Blackwell Tensor Cores support Microscaling formats for 4B inference(nvidia.com)
1 points by buildbot 3y ago|0 comments
707.
by buildbot 3y ago
Still? I would have thought by now they would have switched to something like a Phase One IXH - https://heritage-digitaltransitions.com/dt-ixh-150mp-cultura... I have a broken betterlight I got just to put on my shelf haha.
708.
by buildbot 3y ago
It’s not mentioned in the article, but the Brock Commons that was the tallest timber building is at UBC! I was there as it was built and finished. It’s an 18 story dorm building one of several nearly equally sized dorms (!!) on campus. It w
709.
by buildbot 3y ago
Fair enough - personally, I'd use a digital camera to scan large things anyway, possibly using an XY gantry if needed. Use a monochrome sensor + color filter wheel for extra points. Scanners terrify generations equally in my opinion ;)
710.
by buildbot 3y ago
Scanners are also a tech that has pretty much plateaued - correct me if I am wrong, but the linear arrays are fairly cheap and the resolution is a product of the linear array and the minimum step size, and steppers have not gotten much bett
711.
by buildbot 3y ago
Wormhole does support scale out though via it's built in networking? And there seem to be links for something on top of the board, NVLINK style. And yes, I know, I've been working on LLM pretraining for about 4 years now, since 20
712.
by buildbot 3y ago
FP19 is perfectly fine for training - llama is trained in fp16, many train in bf16, with microscaling exponents, you can do 6 bit training : https://arxiv.org/abs/2310.10537
713.
by buildbot 3y ago
Oh wow, what a good idea. I knew that silicon became transparent around 1050nm, but never thought to use that concept to scan a chip to see through the initial layers! I literally have a 1070nm bandpass filter and a 100MP monochrome camera,
714.
by buildbot 3y ago
Compute limited - an m2 ultra has 27 tflops, a 4090 80+
715.
by buildbot 3y ago
Nice, I tried to use QLoRA+FSDP in the past with litgpt and obviously at that time it did not work. This is very useful!
716.
by buildbot 3y ago
Lit-GPT is what I have been using to pretrain models at home: https://github.com/Lightning-AI/litgpt Using the openwebtext example, I can train a 700M param model to 2.6 loss in a few days on dual 4090s. Pretty awesome
717.
by buildbot 3y ago
eCos! It also ran Sinar emotion digital backs
718.
by buildbot 3y ago
Kinda, it turns out SMT has lots of security pitfalls and having many tiny single thread cores vs. some heavily threaded cores works better in practice. (I love the niagara chips, I had a T1 and T2 box for a bit!)
719.
by buildbot 3y ago
How is a non generally intelligent AI like Gemini know now is a bad time to talk about Indian politics? Are models in general supposed to avoid offending anyone, anytime? Companies align the models so hard to avoid controversial opinions th
720.
by buildbot 3y ago
I would say it’s reasonably accurate to say he’s accused of it given the number of articles and related information that pop up searching for Modi and Fascism - for example, https://en.wikipedia.org/wiki/Hindutva I don

More ›