The corpus embeddings are generated from this compressed JSONL using the specified transformer ML model