Embedding Quantization: 25-45x retrieval speedup, 32x or 4x less memory usagehuggingface.co·4 pts·cubie·0