we made a speech model 13× smaller by making... it repeat itself. 52 layers → 2 shared blocks. 2.04B parameters → 153M. The encoder loops 24 times. The decoder loops 28. Smaller weights, roughly the same compute. Orukshow more

4/ to achieve maximum memory efficiency, we quantize model... weights to ~4-bit, getting the language model under 20GB with room for the kv cache, perception encoder, and drafter alongside it. a dflash drafter proposes blocks of tokens the main model verifies in parallel, so it stays responsive.show more

the way ts loops when you put it on... repeat is so good😭😭😭😭show more

The making of my oil painting of fruit loops

Different radii can cover the same distance because distance... depends on the path traveled, not just the radius itself. When objects move along curves or arcs with different radii, a smaller radius can be compensated by a larger angle, while a larger radius can use a smaller angle. In both cases, the arc length can end up equal-even though the curves look very different.show more

Working on a smaller version for the desk of... the Statue of Satoshi Nakamoto: 19 layers of stainless steel. These are the different sizes: Public venues: 90cm 61 layers Home version: 45cm 31 layers Desk version: 20cm 19 layersshow more

The Myrnohrad pocket gets smaller and smaller with Ukrainian... groups eliminated by targeted strikes.show more

Do 3D reconstruction transformers really need a billion parameters,... or are most of those layers just doing the same thing over and over? Introducing Déjà View: a single transformer block, looped K times, that matches or beats models 8–10× its size with lower compute. 🧵show more

If you guys missed it, we made a smaller... more subtle TADC ad for japan that played on small little pillars outside, here's the full version!show more

Taking a break from the bigger roguelike mod and... making a smaller version. It's called Stock Run. If you've ever played rounds it'll pretty much have the same gameplay loop.show more

In RADDBG, the "Step Out" command has been upgraded... to support block stepping! It now lets you step through blocks, out of loops, or (the traditional behavior) out of function frames.show more

The tops keep looking smaller and smaller with Cat

The smaller the better???

This is the same model as before, but with... a smaller belly before eating—what do you think? 🩶🖤 #VtuberRigger #animation #VTubeStudio #ENVtuber #Esvtuber #live2dcommission #Live2dWIP #Vgenshow more

Day 15 of learning AI bs from scratch >... looked into seq2seq models > understood the encoder-decoder architecture > implemented a basic Seq2Seq model using RNNs > how the decoder generates outputs one step at a time > read couple more chapters of > understood the fixed-context bottleneck I updated the resource list blog, link below, it has more resources organized now. Fell asleep on the desk towards the morning, rweeting rn after waking up.show more

I made the face shadow smaller, now the chin... aligns much better with the ears. Looks better!show more