manmal

27 ms·

661.
by manmal 2y ago
At FP16 you‘d need 64GB just for the weights, and it‘d be 2x as slow as a Q8 version, likely with little improvement. You‘ll also need space for attention and context etc, so 80-100GB (or even more) VRAM would be better. Many people „just“
662.
by manmal 2y ago
Don’t forget to add a lot of extra space if you want a usable context size.
663.
by manmal 2y ago
For parents, most definitely. I have a hard time sitting through a Caillou episode. Bluey is also way more authentic, in that interactions could really happen exactly like that in real life. Compared directly, Caillou seems synthetic.
664.
by manmal 2y ago
> The International Nut and Dried Fruit Council may or may not be a conflict of interest about seed oils. Why do you doubt that there’s a conflict there?
665.
by manmal 2y ago
Have you even read my original comment? I‘m criticizing their internal stance towards the iOS platform, and not the feature set.
666.
by manmal 2y ago
They chose to make an iOS app though, and, it seems to be a second class citizen.
667.
by manmal 2y ago
Not to my knowledge, if you mean the models for the Chinese market. A 10k model in China is also vastly different from a 30k model in Europe. It’s quite an unfair comparison.
668.
by manmal 2y ago
Where do you get these prices from? The cheapest BYD selling in Germany is 30k, as far as I can see. With a 20k vehicle launching sometime in 2025.
669.
by manmal 2y ago
> Some features will appear first on our Android app, and then eventually we’ll add them to the iOS app. This is because the majority of our development team uses Android phones, and generally we’re building things for ourselves, so natu
670.
by manmal 2y ago
Sibling comment already explains the open weights part; even if there was some small binary wrapper involved, it’s trivial to stop if from phoning home.
671.
by manmal 2y ago
What’s the mechanism of dying of cognitive decline? Cognition doesn’t sound vital to bodily functions.
672.
by manmal 2y ago
My deep respect for this decision. There‘s this German saying - you ought to go when things are at their best. The older I grow, the more I‘m believing that quality of life is what we should always optimize for. Not duration, anyway. Or - b
673.
by manmal 2y ago
That’s a problem big grids also share. It‘s probably easier to buffer power for a small neighborhood, than for a city.
674.
by manmal 2y ago
Looking at the interactions in metabolism, I don’t see it. A low B6 diet and focus on conversion to P5P and oxidation of the excess (Aldehyde Oxidase, and maybe by proxy the ALDHs) is a safer bet.
675.
by manmal 2y ago
Parent just forgot a decimal separator I think.
676.
by manmal 2y ago
Maybe you mean B6, which is associated with small fiber neuropathy. B12 has to jump through a number of hoops before it‘s even converted into a usable form [1]. I‘m sure you‘re meaning well, but B12 deficiency is a wide spread problem [2],
677.
by manmal 2y ago
He‘s basically jumping from hype to hype, riding the forefront wave of attention, funneling it into his wallet. I think he‘s extremely intelligent and quick also, both of which are necessary to do this. I wouldn’t be able to.
678.
by manmal 2y ago
The Viennese often add whipped cream on top.
679.
by manmal 2y ago
Where I live, we do say Bodenschwelle. I‘m surprised it’s not a ubiquitous term. Granted, you never talk about these with people when traveling.
680.
by manmal 2y ago
No, Q4 just barely fits, and with a longer context sometimes things freeze. I definitely have to close Xcode.
681.
by manmal 2y ago
My experience with quantizations is that anything below 6 is noticeably worse. Coherence suffers. I’ve rarely gotten anything really useful out of a Q4 model, code wise. For transformations they are great though, eg convert JSON to Markdown
682.
by manmal 2y ago
I‘ve tried the very early Q4 mlx release on an M1 Max 32GB (LM Studio @ default settings), and have run into severe issues. For the coding tasks I gave it, it froze before it was done with reasoning. I guess I should limit context size. I d
683.
by manmal 2y ago
I guess I won’t be needing that 512GB M3 Ultra after all.
684.
by manmal 2y ago
My understanding is that the GPU must still load its assigned layer from VRAM into registers and L2 cache for every token, because those aren’t large enough to hold a significant portion. So naively, for a 24GB layer, you‘d need to move up
685.
by manmal 2y ago
Messages on Mac freezes up for 5-10 seconds every time I get a Screentime request. It’s a hot mess.
686.
by manmal 2y ago
Fast forward and B+ trees end up making sense (again).
687.
by manmal 2y ago
That‘s cosmic. I was wondering about the feasibility of exactly this system today while driving to work. Zettelkasten seems formalized/rule-based enough to work, and the limited context of each Zettel should be ideal for LLMs. I guess
688.
by manmal 2y ago
FWIW, I‘ve seen consensus here on HN in another thread on SQLite-on-server, that there must indeed be a central DB for metadata (user profiles, billing etc), and all the rest is then partitioned.
689.
by manmal 2y ago
Do you mean via distributed transactions? I‘m clueless in that area, and don’t understand how hundreds of self sufficient services would partake in transactions that span different sets of those services, without entangling them badly.
690.
by manmal 2y ago
Is it even realistic to depend on transactional guarantees, with hundreds of services hammering the DB(s) more or less concurrently? Don’t they need to coordinate themselves outside of transactions?

More ›