We're currently using a multilingual sentence-transformers models (https://www.sbert.net/) model to compute text embeddings used for text similarity. It's a great toolkit. I noticed a lot of the models were trained and released a couple years ago and it got me wondering what the best open source models are these days. Looked for alternatively models out of curiosity but didn't have much luck. Have there been new releases with improved accuracy, performance and / or larger context windows?
What are you using for text embeddings?
… for one I would like one with a wider attention window.