Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
72 tokens/sec
GPT-4o
61 tokens/sec
Gemini 2.5 Pro Pro
44 tokens/sec
o3 Pro
8 tokens/sec
GPT-4.1 Pro
50 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

SmoothQuant+: Accurate and Efficient 4-bit Post-Training WeightQuantization for LLM (2312.03788v1)

Published 6 Dec 2023 in cs.LG and cs.CL

Abstract: LLMs have shown remarkable capabilities in various tasks. However their huge model size and the consequent demand for computational and memory resources also pose challenges to model deployment. Currently, 4-bit post-training quantization (PTQ) has achieved some success in LLMs, reducing the memory footprint by approximately 75% compared to FP16 models, albeit with some accuracy loss. In this paper, we propose SmoothQuant+, an accurate and efficient 4-bit weight-only PTQ that requires no additional training, which enables lossless in accuracy for LLMs for the first time. Based on the fact that the loss of weight quantization is amplified by the activation outliers, SmoothQuant+ smoothes the activation outliers by channel before quantization, while adjusting the corresponding weights for mathematical equivalence, and then performs group-wise 4-bit weight quantization for linear layers. We have integrated SmoothQuant+ into the vLLM framework, an advanced high-throughput inference engine specially developed for LLMs, and equipped it with an efficient W4A16 CUDA kernels, so that vLLM can seamlessly support SmoothQuant+ 4-bit weight quantization. Our results show that, with SmoothQuant+, the Code Llama-34B model can be quantized and deployed on a A100 40GB GPU, achieving lossless accuracy and a throughput increase of 1.9 to 4.0 times compared to the FP16 model deployed on two A100 40GB GPUs. Moreover, the latency per token is only 68% of the FP16 model deployed on two A100 40GB GPUs. This is the state-of-the-art 4-bit weight quantization for LLMs as we know.

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (6)
  1. Jiayi Pan (19 papers)
  2. Chengcan Wang (1 paper)
  3. Kaifu Zheng (1 paper)
  4. Yangguang Li (44 papers)
  5. Zhenyu Wang (150 papers)
  6. Bin Feng (44 papers)
Citations (6)
Github Logo Streamline Icon: https://streamlinehq.com
X Twitter Logo Streamline Icon: https://streamlinehq.com