I regularly use vector embeddings in my work, especially for sentence similarity tasks (vector search using cosine) and classification tasks (logistic regression/ NN on embeddings). When I started two years ago, the best model I found was ‘Sahajtomar/french_semantic’ that barely has a 100 downloads.
Every 6 months i check the new SOTA models for sentence similarity and they always perform significantly worse than this one. I am doing something wrong ? I am using every model “out of the box” and I have a neutral benchmark to test my use cases.
It seems weird that i haven’t found a model that perform at least on a similar level …