> - BERT models are smaller and can run with lower latency and serve larger batches with lower vram requirements The most recent version of Wolfram Language (a

> - BERT models are smaller and can run with lower latency and serve larger batches with lower vram requirements The most recent version of Wolfram Language (aka Mathematica) uses by default BERT models for embedding. (Say, for this function: https://reference.wolfram.com/language/ref/CreateSemanticSearchIndex.html https://reference.wolfram.com/language/ref/CreateSemanticSea... .)