International control hasn’t solved the nuclear weapon problem, the best we can do is threaten to invade people who try to develop them and don’t already have them. Once you join the club, you are in and cant be kicked out.
I’ve considered that, but the rules of game theory don't change just because they are inconvenient. China has a lot of talent and a lot of problems that they are banking on automation (along with AI) to solve. They see it as an advantage that they can’t afford for America to monopolize and one they are uniquely suited to lean into (having lots of smart educated people). There is no world where (a) China willingly gives up its edge and (b) trusts America to give up its edge (the reverse is likely true also). And that is only one pair of countries to consider.
Heck, after visiting China this summer, I’m even more worried about an accidental terminator/skynet-style robot apocalypse.
Assuming China makes progress only because they copy the west is really arrogant. And it would require strict international control as you say, a world government that isn't going to happen anytime soon.
https://huggingface.co/Jundot/Qwen3.6-35B-A3B-oQ6-fp16-mtp is my favorite right now (and nothing else is even close), you should be able to get 80-90 tok/sec enabling MTP and disabling turboquant. Supposedly the 6-bit quant is essential for decent MTP performance, although I didn’t test the 4-bit quant.
You need to look at where those nuclear plants are being built, in the east, and where the new energy is out west that makes it appear that nuclear is falling behind. Having power generation that doesn’t need to be transported via UHV lines is useful, it also acts as a nice baseline that keeps the grid stable.
Buyers bid higher because sellers are giving up their low property tax, not because the buyer will have a low property tax in a decade or two. Sellers are hesitant to sell because they are giving up a lot, California had to pass a special law just to let old people downsize without losing their low property tax on the big home they raised a family in.
With rent control, you are limited in how you can increase rent later, so it’s to your benefit to let the apartment sit vacant longer until you get a desirable rent. Without rent control, because you can just increase rent next year by whatever, you can rent it at the market rate right away without waiting.
It’s totally possible, but would seriously need to not use much parallelism, and definitely not going to use the GPU much. So it would be really slow, but possible.
for 3.8? I can get 40 tok/sec on it (M3 Max 64GB), but I don't use it because I can get 90 tok/sec with 3.6 MoE (MTP + 6 bit quant), and I don't notice any quality improvements for my tasks using a dense model.
Did you ask Gemini or DeepSeek to look at your oMLX server log to see what was going on? This can help a lot if it is just a misconfiguration.
Thats interesting, but it seems most subway systems are still QR code based? Anyways, I’ll try to see if I can get that to work next time I go to China.
Thanks, I forgot about. I have an iPhone so didnt have to worry about the incompatibility with Android. If it makes android users feel any better, here in Seattle orca cards can go on android phones but not iPhones.
That isn’t right. Use oMLX or something similar to serve the model with internal MTP enabled. You should get at least 40 tok/sec, but I’m not sure what hardware you are using. Dflash will mess up batching, not really worth it. You need to sift through hugging face for the right model though, and some of the MTP models are meant for rapid MLX and not oMLX.
But you should probably just get a suica in your apple or google wallet, you don’t actually need an IC card. I’m not sure if using a phone makes it less fast though, I couldn’t tell (we had to get a card for my kid since he doesn’t have a phone).
Definitely better than the QR codes used by China.
How is that not just better since the power requirements are less? The whole point against conventional ACs and for what we call heat pumps is purely that they use less electricities (well, they can also heat and cool, so you don’t need two systems). People like heat pumps simply because they are more energy efficient. It isn’t guerrilla marketing making people like a new thing that is the same as the old thing.
> So just imagine, having a normal job in the US and you're asked to leave to renew your visa and this stuff happens.
The only way I can see this happening is if you try to convert from OTP to an H1. I used to do some runs up to Vancouver to help a friend do that.
This just generally sucks for anyone making plans in their life. I remember my wife's greencard interview in Guangzhou, and it would have sucked if they instituted this the day before she did that.
No, I don't care what you think about the devices, and I wouldn't have even bothered replying if it was just criticism. Its comments like "we need to make Apple go away" that piss me off, because you are trying to force your opinion onto everyone else because you don't like that Apple devices exist, hence the "misery loves company" comment. It really pisses me off when people want to take away my choices.
I mean, I can tell you the same thing, to f*ck off. If you don't want to buy Apple products, don't buy Apple products! But don't you dare prevent me from buying them because your butthurt demands that Apple be destroyed because their success makes you uncomfortable.
I don't think the $20/month packages come with API access at all.
I bought a refurbished M3 Max a couple of years ago just to try things out, I get 90 toks/s which is good enough for a lot of tasks. But ya, when DRAM prices come back down I'll definitely look for a better local rig.
Sharding a dense model using Tensor Parallelism (TP) across dual RTX 5090s has a significantly worse performance penalty over PCIe than sharding an MoE model.
If you are using multiple GPUs, MoE is basically going to be your only workable choice unless you can leverage pipeline parallelism (only half your GPUs can work on a prompt at a time, so you need to process prompts back to back in a pipeline setup, and they better be doing similar things because your vram is limited).
> "if your model can fit into the VRAM" can be true for the Mac as well.
It is much more likely for your model to fit in large unified memory of a Mac than the smaller more limited memory of a GPU. Even going with two 5090s, you now have to shard your model and that is a PITA.
But it turns out that MoE is the solution both for running models on macs of limited computer power means (not as fast as GPUs), and on multiple GPUs that require sharding the model.