Is (Selective) Round-To-Nearest Quantization All You Need?
This work addresses the need for efficient and practical quantization techniques for LLMs, offering a viable alternative that could simplify deployment, though it is incremental as it builds on existing RTN methods with selective precision improvements.
The paper tackles the problem of quantizing Large Language Models (LLMs) by challenging the dismissal of Round-to-Nearest (RTN) quantization, showing that RTN can achieve similar accuracy and better token generation throughput than more advanced methods while being cheaper to apply.
Quantization became a necessary tool for serving ever-increasing Large Language Models (LLMs). RTN (Round-to-Nearest) is perhaps the simplest quantization technique that has been around well before LLMs surged to the forefront of machine learning (ML) research. Yet, it has been largely dismissed by recent and more advanced quantization methods that claim superiority over RTN in nearly every aspect of performance. This work aims to dispel this established point of view, showing that RTN is not only much cheaper to apply, but also its token generation throughput can be better than and accuracy can be similar to more advanced alternatives. In particular, we discuss our implementation of RTN based on the recent Marlin kernels and demonstrate how the accuracy of RTN can be gradually improved by selectively increasing the data precision format of certain model layers and modules. Based on our results, we argue that RTN presents a viable and practical choice for quantizing LLMs.