FlattenQuant: Breaking Through the Inference Compute-bound for Large Language Models with Per-tensor Quantization
This addresses deployment bottlenecks for LLMs by reducing latency and memory usage, though it is incremental as it builds on existing quantization techniques.
The paper tackles the compute-bound issue in large language model (LLM) inference by introducing FlattenQuant, a per-tensor quantization method that flattens large channels to reduce tensor values, achieving up to 2x speedup and 2.3x memory reduction with minimal accuracy loss.
Large language models (LLMs) have demonstrated state-of-the-art performance across various tasks. However, the latency of inference and the large GPU memory consumption of LLMs restrict their deployment performance. Recently, there have been some efficient attempts to quantize LLMs, yet inference with large batch size or long sequence still has the issue of being compute-bound. Fine-grained quantization methods have showcased their proficiency in achieving low-bit quantization for LLMs, while requiring FP16 data type for linear layer computations, which is time-consuming when dealing with large batch size or long sequence. In this paper, we introduce a method called FlattenQuant, which significantly reduces the maximum value of the tensor by flattening the large channels in the tensor, to achieve low bit per-tensor quantization with minimal accuracy loss. Our experiments show that FlattenQuant can directly use 4 bits to achieve 48.29% of the linear layer calculation in LLMs, with the remaining layers using 8 bits. The 4-bit matrix multiplication introduced in the FlattenQuant method can effectively address the compute-bound caused by large matrix calculation. Our work achieves up to 2$\times$ speedup and 2.3$\times$ memory reduction for LLMs with negligible loss in accuracy.