coolspot

21 ms·

301.
by coolspot 3y ago
220 Bp * 8 = 1760 Bp 1760 Bp takes ~ 1760 Gb of (V)RAM when 8-bit quantized. Plus you will need some memory for state. So you will need at least 1760/24=74 consumer-grade GPUs (3090/4090) or 1760/80=22 professional-grade GPUs
302.
by coolspot 3y ago
Magic mouse has one physical clicker. To imitate “right click” you have to lift fingers from the left side and click. They can’t make a mouse with two physical clickers.
303.
by coolspot 3y ago
Ah, yes.
304.
by coolspot 3y ago
Same reason why Mac mouse doesn’t have right click and why ~double-click~ [edit: pressing “return”] on a folder is renaming, and why window controls are on a left side of a window and do things unlike MS Windows (not closing, not minimizing
305.
by coolspot 3y ago
“ There are many exciting use cases for generative speech models, but because of the potential risks of misuse, we are not making the Voicebox model or code publicly available at this time. While we believe it is important to be open with t
306.
by coolspot 3y ago
Real reason: he ran out of diet Coke.
307.
by coolspot 3y ago
It is much easier to transfer X amount of heat with a liquid than with air. Liquid travels in pipes and has enormous mass heat capacity (Per unit of volume, water has about 3200 times the specific heat capacity of dry air (at 77°F)). Air ne
308.
by coolspot 3y ago
We are talking about the CEO who edited user comments that criticized him: https://www.huffpost.com/entry/reddit-ceo-edits-user-comment...
309.
by coolspot 3y ago
Ep 1315
310.
by coolspot 3y ago
That’s a very nice house!
311.
by coolspot 3y ago
It may be presented as seamless unified memory, but it isn’t. Underlying framework still has to figure out how to allocate your data to minimize cross-unit talk. Each unit has independent CPU, GPU and (V)RAM, but units are interconnected vi
312.
by coolspot 3y ago
If we are talking about indexing, writing M$ is easier to find in an index because it is a such unique token. MS can mean many things (e.g. Miss), M$ is less ambiguous.
313.
by coolspot 3y ago
Same, but they don’t care. These posts are aimed at potential investors who care mostly about returns and financial track record.
314.
by coolspot 3y ago
Recently announced Oculus Quest 3 has full color and depth AR pass-through for $499
315.
by coolspot 3y ago
> Does someone really sit on their couch, put on a massive headset, and scroll through their vacation photos? A male sitting alone in a dimmed room watching photos with a smug on his face. I think they were hinting at something else than
316.
by coolspot 3y ago
That’s what llama.cpp does, including offload to a disk. It allows you to run models as big as any combination of your VRAM, RAM or disk. But in the end, if it doesn’t fit into GPU VRAM, it will be slow. For example Guanaco-33B generates ~
317.
by coolspot 3y ago
4090 is twice faster than 3090 in almost all metrics.
318.
by coolspot 3y ago
Guanaco is indeed very capable and can replace GPT 3.5 in almost all scenarios, based on my tests. Easy way to self-host it is to use text-generation-webui[1] and 33B 4-bit quantized GGML model from TheBloke[2]. [1] https://githu
319.
by coolspot 3y ago
Hacker News. You all are my community!
320.
by coolspot 3y ago
Just use hckrnews.com , it shows frontpage posts from previous days.
321.
by coolspot 3y ago
Guanaco 33B ELI5: > ELI5 ML model quantization > Here is a simpler explanation of what AI model quantization means: Imagine you have a big box full of marbles, where each marble represents a piece of information. Each marble has a uni
322.
by coolspot 3y ago
What case/MB/GPUs do you use for your dual 3090 build? Liquid cooled cards?
323.
by coolspot 3y ago
I would presume that any existing TPUs in Google Cloud were already doing something for Google or its tenants.
324.
by coolspot 3y ago
> But around 2000, everyone and their mother was talking about how MSFT can't produce any software beside complete garbage. > None of that was true. Ummm… In 2000s MSFT was producing complete garbage. Internet Explorer, Windows X
325.
by coolspot 3y ago
I would love to see napkin math on how much new hardware capacity is needed to serve Bard as a general availability product. Where they got so many TPUs on a such a short notice?
326.
by coolspot 3y ago
They are also testing in West LA.
327.
by coolspot 3y ago
Meatbags are inefficient ancestors that are not needed on the planet anymore. Small population (say a 1000 bodies) will be kept around in a designated area. Rest will be culled to save planet’s resources.
328.
by coolspot 3y ago
> And the rest of the world was better off for it. Except the single mom in a studio apartment trying to get some pay from her art gigs.
329.
by coolspot 3y ago
No one has created even something closed-source that is equal to GPT4.
330.
by coolspot 3y ago
https://github.com/Genymobile/scrcpy/issues/36#issuecomment-...

More ›