Deep Learning Models via WebGPU
Hello,
We're currently working on a demo that enables Stable Diffusion, a deep learning model that generates high-quality images according to text prompts, on web browsers with only client-side GPUs. In our use cases, WebGPU is a perfect fit. You can find our GitHub repository for this project here: https://github.com/mlc-ai/web-stable-diffusion.
However, at the moment, it's only possible to run the demo via Chrome Canary (Chrome's nightly build). We're very interested in getting it to work on Firefox as well, but we're encountering vague error messages in Firefox's nightly build with the `dom.webgpu.enabled` flag turned on.
I would love to ask two questions: 1) How mature is Firefox's WebGPU support on Metal? 2) Is there any pointers we could follow to develop WebGPU demos for Firefox?
Thanks a lot!
Chosen solution
Hi, it doesn't start because of bug 1814101 (regression in v.111).
You can find interesting these meta-bugs: 1616739, 1602129.
For communication with devs about WebGPU spec and tests, join the #WebGPU:matrix.org room on Matrix.
Skaityti atsakymą kartu su kontekstu 👍 2All Replies (2)
Chosen Solution
Hi, it doesn't start because of bug 1814101 (regression in v.111).
You can find interesting these meta-bugs: 1616739, 1602129.
For communication with devs about WebGPU spec and tests, join the #WebGPU:matrix.org room on Matrix.
Thanks @TyDraniu for the pointers! Yes, this is exactly the error message I noticed. Happy to watch this ticket and try out again once it is fixed.