← All models · Multilingual MiniLM
Wild · feature-extraction · WASM
Something you couldn't do before this ran in a browser: dump a heap of sentences in a jumble of languages and cluster them by topic. A monolingual model would just sort them into language piles. This one groups the French, Japanese and English sentences about the same thing together — the clusters ignore language entirely.
One sentence per line, languages mixed. The default set has three topics (weather, cooking, football) each written across several languages, deliberately shuffled.