31–40 of 96 posts
Earlier quoted context omitted.
That is quite surprising given that a Pentium 166MHz had zero issues playing mp3s in winamp, 1998 hardware.
More like 1995 hardware.
Earlier quoted context omitted.
Serious question - why? Quake 1 is 25 years old game. It ran good on first Pentium. Apple Watch CPU is much more powerful. I’m not saying it’s not cool. But it should be fast.
A game similar to Quake coded in a modern high level language would probably be too slow on that watch.
In fact, there's a recent game called HOAT or something. Built by a single guy using the Quake engine. Had to return it because it was running with 30fps on my machine.
I was half expecting this to have gyro controls so you could look like a madman while "arm looking". Really neat, in any case.
Now I know what I’m doing tonight.
Quoted post unavailable.
Quoted post unavailable.
I have some experience with ML and I have no idea what you're talking about. It kinda sounds like neural architecture search and sparse models created using weight pruning. Lots of people are working on both of those things, but IMO the latter (if that's what you mean by "Can you add weights non uniformly?") is a dead end for most use cases where you have some sort of accelerator or deep learning instructions availab…
There is an idea here, and it’s a mistake to dismiss it out of hand. Adding weights non uniformly during training (not after) is the key to smaller models that outperform present day GPT3.
A sketch of the algorithm is to start with a 2x2 block of weights, sum the gradients across 10 training steps, then subdivide the quadrant with the highest delta.
Doing this recursively is prohibitive, which is where megatexture comes in.
Many advantages. At runtime you don’t need weight compression because you can simply switch to a lower miplevel if running on a phone. Different accelerators during training can focus on different areas of the network. Weight dropout is an automatic feature. Etc.
If you’ll excuse me, it’s back to hugging the porcelain bowl.
Quoted post unavailable.
Quoted post unavailable.
Network compression is already a thing that is studied and forms of it are already used in production neural net models where latency / cost is important. The way pruning works is not like how a "megatexture" works.