Explain quantization. What are the trade-offs between model size, speed, and accuracy?