web-stable-diffusion VS memory64

Yup, here's their web stable diffusion repo: https://github.com/mlc-ai/web-stable-diffusion

The input is a model (weights + runtime lib) compiled via the mlc-llm project: https://mlc.ai/mlc-llm/docs/compilation/compile_models.html

The MLC team got that working back in March: https://github.com/mlc-ai/web-stable-diffusion

Even more impressively, they followed up with support for several Large Language Models: https://webllm.mlc.ai/

[https://github.com/mlc-ai/web-stable-diffusion

You can run it directly in the browser with WebGPU, https://mlc.ai/web-stable-diffusion/

Using the amazing work of https://mlc.ai/web-stable-diffusion/ I've got the code moved into a Web Worker and running fully local client side. It does require 2GB's of model files be downloaded (automatically), and takes a few minutes for the first load, but it works and once it's going it only takes 20s to make a 512x512 image.

The Apache TVM machine learning compiler has a WASM and WebGPU backend, and can import from most DNN frameworks. Here's a project running Stable Diffusion with webgpu and TVM [1].

Questions exist around post-and-pre-processing code in folks' Python stacks, with e.g. NumPy and opencv. There's some NumPy to JS transpilers out there, but those aren't feature complete or fully integrated.

[1] https://github.com/mlc-ai/web-stable-diffusion

A completed implementation of memory64 for memory-hungry applications.

Indeed, webassembly is moving extremely slowly. I started a project years ago expecting https://github.com/WebAssembly/memory-control/blob/main/prop... and https://github.com/WebAssembly/memory64 to be fixed at some point. Neither are yet, and the project still suffers from it to this day.

I think wasm is still great without these fixes, but I have lost confidence in the idea that wasm will reach its full potential any time soon.

It's in the works: https://github.com/WebAssembly/memory64

Starting with 32bit had some performance advantages because 64bit runtimes can use virtual memory shenanigans to implement bounds checking with zero overhead. In wasm64 they'll have to do explicit bounds checking instead.

Right - currently, everything runs using WASM (32-bit, with 64-bit coming soon [1,2]), and I plan to add support for WebGPU soon!

(WebGPU is the successor to WebGL, which is coming out in April 2023 [3])

[1] https://github.com/WebAssembly/memory64/issues/36#issuecomme...

It shouldn't be a big surprise that a 64-bit pointer extension is out there and being worked on. The great thing about a VM is you can integrate major changes like this when they are needed and with the benefit of experience and hindsight. If the 4GB limit turns out to be restrictive then it can be lifted.

The goal of this proposal is to increase the amount of memory that canisters can access [eventually] bound only by the actual capacity of the subnet. Since, the Memory64 proposal is not standardized 1 yet and its implementation 1 in Wasmtime is not production ready yet, this proposal enables the increase by introducing a new stable memory API.