For sure they quickly move to q8, the output quality difference to bf16 is small compared to the speed/capacity gain

NineStarPoint 5 hours ago | parent | next [-]

Yeah q8 made so littler difference back when I was testing such things I'd be surprised if people could quickly notice that as a change. It's got to be either further quantized or some other type of optimization that kicks in when people notice the drop.

▲

selectodude 5 hours ago | parent [-]

NVFP4 would buy them a huge increase in capacity but I think it would be noticeable.

▲

Caracas288 4 hours ago | parent [-]

Why doesn't someone just try to measure this next time!?

	▲	embedding-shape an hour ago \| parent [-]
		Can't really measure without being sure you aren't being messed around with, when it's a remote platform. Stupidly easy to detect when people run such benchmarks/tests against you as well.

▲

torginus 4 hours ago | parent | prev | next [-]

Some people here have remarked previously that while reduced precision doesn't show up in quick prompts, it does severely impact these models' ability to perform long running tasks - to the point that running these big models with severe quantization might be counterproductive as smaller but less quantized ones perform better.

▲

nonethewiser 4 hours ago | parent | prev [-]

Could this explain Opus?