Birch-san

5 ms·

1.
by Birch-san 1y ago
“Windows Subsystem for Linux” is probably the most confusing example of this (an environment subsystem which provides a Linux userspace to a Windows NT kernel). more intuitive would be to call it a Linux Subsystem for Windows, but presumabl
2.
by Birch-san 2y ago
I think they meant scrollback. as conventionally a backlog would evoke "work yet to be done", whereas in this context we're talking about a conversation history one can revisit.
3.
by Birch-san 2y ago
feels like it could be nice to abide by the license terms https://bria.ai/bria-huggingface-model-license-agreement/ > 1.1 License. > BRIA grants Customer a time-limited, non-exclusive, non-sublicensable, pers
4.
by Birch-san 3y ago
> Is it super resolution? nope, we don't do Imagen-style super-resolution. we go direct to high resolution with a single-stage model.
5.
by Birch-san 3y ago
FID doesn't reward high-resolution detail. the inception feature size is 299x299! so we are forced to downsample our FFHQ-1024 samples to compute FID. it also doesn't punish poor detail either! this advantages latent diffusion,
6.
by Birch-san 3y ago
the FFHQ-1024 examples shouldn't be blurry. you can download the originals from the project page[0] — click any image in the teaser, or download our 50k samples. the ImageNet-256 examples also aren't typically blurry (but they are
7.
by Birch-san 3y ago
cross-attention doesn't need to involve NATTEN. there's no neighbourhood involved because it's not self-attention. so you can do it the stable-diffusion way: after self-attention, run torch sdp with Q=image and K=V=text. I tr
8.
by Birch-san 3y ago
ah, originally lstein/stable-diffusion? yeah that was an important fork for us Mac users in the early days. I have to confess I've still never used a UI. :) this year I'm hoping for efficiency and small models! even if it&#x2
9.
by Birch-san 3y ago
regarding ControlNet: we have a UNet backbone, so the idea of "make trainable copies of the encoder blocks" sounds possible. the other part, "use a zero-inited dense layer to project the peer-encoder output and add it to t
10.
by Birch-san 3y ago
I'm one of the authors; happy to answer questions. this arch is of course nice for high-resolution synthesis, but there's some other cool stuff worth mentioning.. activations are small! so you can enjoy bigger batch sizes. this is
11.
by Birch-san 4y ago
Most things require workarounds, some things aren't possible (or we haven't found workaround yet) and it's not as fast as CUDA. But stable-diffusion inference works, and so does textual inversion training. I was also able to
12.
by Birch-san 4y ago
It's probably the same thing prestodb encountered: https://github.com/prestodb/presto/issues/8993
13.
by Birch-san 4y ago
For performance reasons, glibc may not return freed memory to the OS. You can increase the incentive for it to do so, by reducing MALLOC_ARENA_MAX to 2. https://github.com/prestodb/presto/issues/8993
14.
by Birch-san 4y ago
stable-diffusion supports k-diffusion just fine on M1. You just have to detach a tensor in to_d() to stop the values exploding to infinity. https://twitter.com/Birchlabs/status/1563622002581184517?s=2...
15.
by Birch-san 4y ago
9.3 secs if you sample for 7 steps. 11 secs if you sample for 8 steps. Heun sampler, M1 Max. https://twitter.com/Birchlabs/status/1564792349221330944?s=2...
16.
by Birch-san 4y ago
Neither of these should take minutes. Try Heun sampler, 8 steps, Karras noise schedule. Should be possible to get good images in 11 secs (or 10 secs if you go down to 7 steps). measurements admittedly from M1 Max. https://twitter
17.
by Birch-san 4y ago
About 10 secs per image on M1 Max with the right noise schedule and sampler. https://twitter.com/Birchlabs/status/1565029734865584143?s=2...
18.
by Birch-san 4y ago
You might be eligible to use Google's TPU Research Cloud for free, provided you publicize your results? https://sites.research.google/trc/about/ Otherwise, perhaps you could ask LAION on #compute-allocation?
19.
ARM-native launcher for Live2D Cubism Editor (unofficial)(twitter.com)
1 points by Birch-san 5y ago|0 comments
20.
by Birch-san 5y ago
Even without mods, Minecraft supports programming mechanisms (albeit not robots). redstone blocks can be used like logic gates, to construct traps, play music, open doors or build factories. People have implemented ALUs in redstone.
21.
by Birch-san 5y ago
yes, that demo has very impressive shaders. it models the physics in a more simple way though (i.e. as waves), rather than as thousands of particles, so it can't generalize to more complex scenarios.
22.
by Birch-san 5y ago
network inspector says 2.1MB. but that's dominated by a 1.3MB image. the main assets of the library are: - Box2D.simd.js (422kB) - Box2D.simd.wasm (266 kB) a minimal demo that uses the library can be created in just a few kB: https:&#
23.
by Birch-san 5y ago
it's all modelled using physical laws; see slides 16 onwards of the "Inside Liquidfun" presentation: http://docs.google.com/presentation/d/1fEAb4-lSyqxlVGNPog3G1...
24.
by Birch-san 5y ago
okay, have another look now[0]. I've updated it to use a consistent number of particle iterations regardless of how much time is being simulated. in fact, you've caught that it was using as many as 16 iterations (my intention was
25.
by Birch-san 5y ago
yes, I think so! some examples here[0][1]. [0] https://www.youtube.com/watch?v=2Bw5f4vYL98 [1] https://www.youtube.com/watch?v=atcKO15YVD8
26.
by Birch-san 5y ago
yeah, I've played around with a few approaches for running the timestep and for some reason I don't feel like I get the same results as liquidfun.js. their loop [0] is pretty simple; it's scheduled by `requestAnimationFrame`,
27.
by Birch-san 5y ago
wow! thanks for your contributions. :) would you mind giving some background on how LiquidFun came into existence? is it a 20% time project, or a weekend thing made by Googlers? was there ever a business case for it?
28.
by Birch-san 5y ago
yes. the fastest "regular JS" version achievable would be via asm.js, and WebAssembly can go faster[0]. you can compare the performance of this demo's Wave Machine mode against the one used on the liquidfun.js frontpage[1]: i
29.
by Birch-san 5y ago
Author here. This demo showcases liquidfun-wasm[0], my effort to revive liquidfun[1] (a fork which adds fluid simulation and soft-body physics to Box2D[2]). to make liquidfun-wasm, I repurposed my existing box2d-wasm[3] and pointed it at a
30.
by Birch-san 5y ago
yes, I compiled with -msimd128 to enable LLVM's auto-vectorization. I distribute both SIMD and non-SIMD, and the entrypoint picks whichever distribution your browser supports. for box2d-wasm, SIMD acceleration resulted in a 0.6–0.9% pe

More ›