AFPQ: Asymmetric Floating Point Quantization for LLMs
This work solves memory and bandwidth limitations for LLM deployment, offering an incremental improvement in quantization techniques.
The paper tackles the challenge of deploying large language models (LLMs) by addressing poor performance in floating-point (FP) quantization at small group sizes or sub-4 bits, proposing asymmetric FP quantization (AFPQ) that sets separate scales for positive and negative values, which leads to large accuracy improvements and can be integrated with methods like GPTQ and AWQ without extra storage.
Large language models (LLMs) show great performance in various tasks, but face deployment challenges from limited memory capacity and bandwidth. Low-bit weight quantization can save memory and accelerate inference. Although floating-point (FP) formats show good performance in LLM quantization, they tend to perform poorly with small group sizes or sub-4 bits. We find the reason is that the absence of asymmetry in previous FP quantization makes it unsuitable for handling asymmetric value distribution of LLM weight tensors. In this work, we propose asymmetric FP quantization (AFPQ), which sets separate scales for positive and negative values. Our method leads to large accuracy improvements and can be easily plugged into other quantization methods, including GPTQ and AWQ, for better performance. Besides, no additional storage is needed compared with asymmetric integer (INT) quantization. The code is available at https://github.com/zhangsichengsjtu/AFPQ.