Short write-ups of things that cost me an hour, so they don’t cost me another one later.
Strata runs a 125-billion-parameter model on a PC with a 12 GB graphics card. What it does, why RAM matters more than you’d think, and which budget setups are worth it, from a RAM upgrade to a used RTX 3090 or two retired datacenter cards.
A short tutorial on running Hugging Face models in a web page with no server: the pipeline() function, WebGPU or WebAssembly, Web Workers, and a list of models worth trying, from background removal to a small chat model. My Transcriber app is the worked example.
A short tutorial on Emscripten: install it, build a real C++ project with it, call the C++ from JavaScript, and serve the result. The example is whisper.cpp, OpenAI’s Whisper speech model in plain C++.