Rendering in this context means producing an element tree, or a textual description of one, which is capable of firing off many little (slow) allocations. The implication being that this will not scale to a large number of nodes without slogging the machine unduly.
If it's in memory and off-screen, i think it won't be that slow even for many elements. The actual issue might be the garbage collection pauses, which might become noticeable.
It works the other way too! People don't know that 3 or 2 bits can still be good because they've read the hivemind's opinion that <4bit is unusable trash and they've never experimented themselves. The reality is a lot more nuanced than that, especially for models as large as 284B.
It also depends on your inference setup too. Like I had looping problems on an IQ3_S quant that were actually down to problems with the harness and the inference engine itself, not the quant. So, nowadays when I see people make those types of claims, I think there's potentially some truth to them that the quant is the problem, but I've experienced that other parts of the stack can be equally as responsible for that and it might not actually be the quant.
That’s an intractable number that you could cook up but it’s just busy work in my opinion and the premise that everything that is obviously important needs a revenue number is wrong.
True, but there are so many bad designers, working at famous places. The Liquid Glass person is the perfect example. So many well documented issues with unreadability with that.
Maybe, I agree that Liquid Glass is not nice, but staying always in the realm of the “nice but safe” is also not so great imo. Time simply preserves the good stuff and forgets the bad. Same with flat UI that came with iOS7 (?) - it was not so nice initially, but eventually contributed to the overall aesthetic direction of the whole industry.
Why not just just have most of it in wasm? There are wasm runtimes that don’t require js, and I’m sure Google can adapt their engine. Then they could integrate the other side with rust
You can always compile to wasm, or ship a wasm decoder. You can even ship a JS decoder, but speed and resource consumption are important, ruling both out.
That would likely require copying the image data in and out of the wasm instance's memory, while a pure Rust solution could have safety with less overhead.
Came up with a good list for my project
reply