WasmDNN

WasmDNN is a deep neural network library for generating machine learning models in WebAssembly. The library is written in C++, but an alternative JavaScript/WebAssembly version is also available by compiling the project using Emscripten.

Architecture

Pipeline

Above is an overview diagram of the WasmDNN library pipeline using an example model designed to operate on the MNIST dataset. To configure this model, one can either do this in C++ or in WebAssembly using JavaScript bindings. In the C++ option, the model configuration is done using the WasmDNN library. In the JavaScript bindings, the model configuration is done using the WebAssembly version of the library compiled, together with Wasm++ and WABT, using Emscripten. Both model configuration options are equivalent and should be capable of generating identical machine learning Wasm model to train, test and predict on the MNIST dataset.