Lightning Talk: Accelerate WebAssembly AI with Backend Graph Compliers on a Wide Variety of AI HW Accelerators
summary
Prototype WebAssembly runtime injects a transparent backend interposer that intercepts ML framework calls (it hijack load_model/predict) and redirects models into mainstream graph compilers (TVM, XLA, OpenVINO et al.) so WASM binaries run on CPU, GPU, FPGA, or other AI accelerators without source changes. Implementation builds graph-compiler libraries into the runtime, converts models to compiler IR at load time, runs inference via the compiler backend, and demonstrably executes end-to-end demos from cloud to edge with improved accelerator utilization.
AI-written summary · about
related talks
32mEmpower WebAssembly ML Over Diverse AI Hardware Accelerators Outside the Browser
37mGetting Started with AI and WebAssembly
37mAI/ML at the Extreme Edge with WebAssembly: A Path Forward
13mLightning Talk: WebAssembly from the Inside Out
25mLeveraging Wasm for Portable AI Inference Across GPUs, CPUs, OS & Cloud-Nativ...
31mRunning AI workloads with WASM - is it production ready? [Panel]
42mWebAssembly? Web and Assembler? WTF!
08mLightning Talk: Build trusted cloud apps with Wasm: WebAssembly Micro Runtime is ready