The last mile of a long road: faster NumPy in the browser
Summary
The post explains how NumPy can run in the browser with accelerated BLAS via OpenBLAS in WebAssembly using Emscripten-forge. It presents benchmark results showing large speedups (e.g., 30.92× for float32 matmul and 14.90× for float64 at n=1024), discusses the packaging model and upcoming SIMD improvements, and describes the impact on browser-based scientific computing with environments like JupyterLite and Notebook.link.