Deploying machine learning in wasm to CPU, GPU and NPU with the new wa...
summary
A new wasi-nn WinML backend lets wasm apps run ONNX models on CPU, GPU (DirectML/DirectX) and NPU paths by mapping wasi-nn APIs to Windows ML calls, implemented in Rust with windows-rs. Example runs MobileNet.onnx image classification end-to-end (preprocess → set_input → compute → get_output) and returns top-1 cat at ~96% probability on GPU and similar on CPU. Known limits: FP16 experimental (internal buffers still FP32) and NPU support pending (PR #8756).
AI-written summary · about
related talks
22mMachine Learning with Wasm (wasi-nn)
33mBringing AI to the Heterogeneous Edge with WebAssembly & ONNX
14mWasm, WebGPU, & WebNN: How compute abstraction are enabling client side AI
33mEfficient and Cross-Platform LLM Inference in the Heterogenous Cloud
25mPathfinder to GPU Offload in WASM
18mWeb AI on next generation AI PCs
33mWebAssembly based AI as a Service with Kubernetes
26mMachine Learning in Fastly's Compute@Edge