I tried it and the experience was not great. I first enabled chrome://flags/#force-enable-webgpu-interop which did nothing. Next, I enabled chrome://flags/#enable-unsafe-webgpu which made some WebGPU demos work, but they all use my CPU's integrated GPU, which is worse than using no GPU acceleration at all. Next, I found out that there is a powerPreference: 'high-performance' option. It can be enabled with chrome://flags/#enable-webgpu-developer-features and I can now request a GPU with 'high-performance' power preference, although there is no way to check whether this actually selected the correct GPU, and it is useless to do it like that anyway, since no website actually sets this (experimental) option. Fortunately, there is the flag chrome://flags/#force-high-performance-gpu which should solve this, but it is not available for my platform. I have a fairly standard RTX 3060, but apparently, the most common GPU (according to the steam hardware survey) is not supported.

Unfortunately, Firefox does not work any better. After setting the flag dom.webgpu.enabled to true in about:config, a few WebGPU examples (e.g. https://webgpu.github.io/webgpu-samples/?sample=helloTriangl...) work, but many other examples crash the browser. And of course, Firefox can't select the correct GPU either.

We use the ONNX runtime for small models in the browser https://github.com/microsoft/onnxruntime

Transformers.js supports a lot of models since v4

There's, quietly, a llama.cpp WebGPU backend that works *great*.

Some hacking required, it's unsupported, a side project for one of the lead maintainers and someone in school.

Note of caution, llama.cpp isn't what it was, the grunt-level maintainers are left to their own devices. There's one key subsystem where things break regularly and the engineering is poor, and the "lead maintainer" is aggro and isn't really involved after delivering their big refactor that was DOA, other than telling people on issues he's never seen their repro himself. Carefully pick models you can support down to "I can patch around the Jinja template engine". Go through the same llama web server APIs. c.f. github / telosnex / fllama if you need a reference

(n.b. seeing peer comments, its much better than ONNX, ONNX never got within spitting distance of llama.cpp, my understanding from watching the LLM runtime is its used for Windows AI features so their models probably work great on Windows x Qualcomm. ONNX is great for smol models though, like VAD, a god send even.)