@jiayq The baseline compared against here is what you find from hosted SDXL services. We also have <5s results available but at a higher price point if that’s what you’re interested in.
@OctoML Profiler uses the latest PyTorch 2.0 compiler technology to give developers insight into ML model performance, optimize deep learning applications, and right-size hardware choices without leaving the dev environment!
Check it out 👉 https://t.co/c1hCTyOsNy
Stable diffusion models are fun and usually take a server to run. Today we are happy to share the Web Stable Diffusion project, which brings SD models from @huggingface and @pytorch to browser clients through @WebGPU support. Checkout https://t.co/9av529Ol7X for our demo!
We just introduced PyTorch 2.0 at the #PyTorchConference, introducing torch.compile!
Available in the nightlies today, stable release Early March 2023.
Read the full post: https://t.co/DFO0HojkMt
🧵below!
1/5
@bernhardsson@brendandburns pushed this concept in “Metaparticle” exploring this concept but it seemed to lose steam. Would love to know why? https://t.co/BVXnIVhfc8
@bernhardsson Interesting historical discussion about taking
@deno_land's VM level permissions and extending them to per-library/module. Still NotImplemented, but V8 VM level is definitely better than nothing:
https://t.co/JM8l8gv2ZK
Minimalism for the mind. A fantastic read that I'm now looking forward to digesting for the rest of the weekend. h/t to @coffeephoenix
Subtraction, Not Addition, by @ibringtraffic https://t.co/E6ZsVcUaFX
@karlhigley Haven’t used it myself, but I wonder if there’s a Feast library component that would serve your need of decoupling your source and sink in a principled way: https://t.co/M66xiW8ThN
Best way to ring in the new year! @erika_alden_d and I are excited to share our story on high-throughput continuous evolution, available now in @naturemethods. We've added a ton to the story since the preprint, and hope you'll check it out! thread 1/n... https://t.co/e76mfVIlNN
Some of the coolest work I have seen in a while. Now we just need to compile these continuous neural CAs into genetic regulatory networks for some synbio breakthroughs.
https://t.co/z17Z5nfEjb <- one and only Distill. A nice introduction to how neural CA works and how it may be a potentially useful model of morphogenesis and regeneration processes in developmental biology.
@zzznah
2/
@migueldeicaza@petewarden@roeschinc Correct, TVM has pretty sophisticated/flexible fusion capabilities. Though there are opens around mixing custom TVM ops into the rest of your framework, though we can always compile the entire network at the end of the day to get around that. :)