For embeddings, it may be overkill. Smaller BERT-type models can provide good embeddings when fine tuned with a contrastive learning objective. Eg: https://sber

For embeddings, it may be overkill. Smaller BERT-type models can provide good embeddings when fine tuned with a contrastive learning objective. Eg: https://sbert.net https://sbert.net.