Train 400x faster Static Embedding Models with Sentence Transformers

This blog post introduces two new efficient static embedding models for English retrieval and multilingual similarity tasks. The models run 100-400x faster on CPU than state-of-the-art alternatives while retaining 85%+ performance. Key features include contrastive learning, Matryoshka representation, and extensive training datasets, resulting in models that enable on-device and low-power applications.