📄 中文摘要
大型语言模型（LLMs）在规模和能力上不断增长，如何高效部署成为一大挑战。量化是一种强大的模型压缩技术，通过降低神经网络中数值（权重和激活）的精度来减少模型大小并加速推理，而不会显著影响性能。具体而言，参数可以用16位、8位甚至4位整数等低精度格式代替32位浮点数。这种简单的数值转换显著改善了内存使用、计算效率，并提高了在资源受限硬件上的部署可行性。
缩小数字，而非能力：理解大型语言模型中的量化
📄 English Summary
Shrinking Numbers, Not Power: Understanding Quantization in Large Language Models
As Large Language Models (LLMs) continue to expand in size and capability, efficient deployment poses a significant challenge. Quantization emerges as a powerful model compression technique that reduces the precision of numerical values (weights and activations) in neural networks. Instead of using 32-bit floating-point numbers, parameters can be represented in lower-precision formats such as 16-bit, 8-bit, or even 4-bit integers. This straightforward numerical transformation leads to substantial improvements in memory usage, computational efficiency, and deployment feasibility on resource-constrained hardware.