vigilo-wasm-gpu
The GPU build of vigilo-wasm: the same in-browser proctoring engine, with model inference sent to WebGPU first and WebAssembly SIMD as the fallback.
188 downloads / 30 days · npm
npm install vigilo-wasm-gpuvigilo-wasm-gpu is the GPU-accelerated build of vigilo-wasm, published from the repository's gpu branch. The Rust engine, the fusion rules and the public API are the same. What changes is where the models run.
Every inference session is created with ['webgpu', 'wasm'] as its execution providers, so onnxruntime-web puts the convolutions on the client's GPU and drops back to WebAssembly SIMD on browsers or machines without WebGPU. The package works either way; the GPU just makes it faster where one is available.
The biggest win is gaze. At 448×448 the gaze model costs more than face and pose combined on the CPU path (104 ms warm), which is why the CPU build only runs it every other frame. Moving the heavy convolutions to the GPU is where the branch documents its 5 to 10x speedup.
Everything else carries over from the CPU build: frames stay in the tab, identity checking is left out, a hidden tab is logged as degraded time, and fusion replays deterministically from recorded signals.