The case for 4-bit precision: k-bit Inference Scaling Laws

TL;DR

This paper develops inference scaling laws for large language models to identify the optimal bit-precision, finding that 4-bit quantization is nearly universally optimal for balancing model size and zero-shot performance.

Contribution

The study provides extensive empirical analysis of quantization effects on LLMs, establishing that 4-bit precision offers near-optimal trade-offs across various model sizes and architectures.

Findings

4-bit precision is nearly optimal for zero-shot performance and model size.

Small block size and quantization data type influence scaling improvements.

Scaling laws guide optimal bit-precision choices for LLM inference.

Abstract

Quantization methods reduce the number of bits required to represent each parameter in a model, trading accuracy for smaller memory footprints and inference latencies. However, the final model size depends on both the number of parameters of the original model and the rate of compression. For example, a 30B 8-bit model and a 60B 4-bit model have the same number of bits but may have very different zero-shot accuracies. In this work, we study this trade-off by developing inference scaling laws of zero-shot performance in Large Language Models (LLMs) to determine the bit-precision and model size that maximizes zero-shot performance. We run more than 35,000 experiments with 16-bit inputs and k-bit parameters to examine which zero-shot quantization methods improve scaling for 3 to 8-bit precision at scales of 19M to 176B parameters across the LLM families BLOOM, OPT, NeoX/Pythia, and GPT-2. We find that it is challenging to improve the bit-level scaling trade-off, with the only improvements being the use of a small block size -- splitting the parameters into small independently quantized blocks -- and the quantization data type being used (e.g., Int vs Float). Overall, our findings show that {4-bit} precision is almost universally optimal for total model bits and zero-shot accuracy.