Train 400x faster Static Embedding Models with Sentence Transformers