LLM-FP4: 4-Bit Floating-Point Quantized Transformers

TL;DR

This paper introduces LLM-FP4, a novel 4-bit floating-point quantization method for large language models that improves performance and flexibility over traditional integer-based quantization, enabling efficient deployment of quantized LLMs.

Contribution

The paper presents a new 4-bit floating-point quantization approach for weights and activations in LLMs, including a strong baseline and a novel per-channel activation quantization technique.

Findings

Quantizes LLaMA-13B to 4-bit with minimal performance loss

Achieves 63.1 accuracy on zero-shot reasoning tasks

Outperforms previous state-of-the-art by 12.7 points

Abstract

We propose LLM-FP4 for quantizing both weights and activations in large language models (LLMs) down to 4-bit floating-point values, in a post-training manner. Existing post-training quantization (PTQ) solutions are primarily integer-based and struggle with bit widths below 8 bits. Compared to integer quantization, floating-point (FP) quantization is more flexible and can better handle long-tail or bell-shaped distributions, and it has emerged as a default choice in many hardware platforms. One characteristic of FP quantization is that its performance largely depends on the choice of exponent bits and clipping range. In this regard, we construct a strong FP-PTQ baseline by searching for the optimal quantization parameters. Furthermore, we observe a high inter-channel variance and low intra-channel variance pattern in activation distributions, which adds activation quantization difficulty. We recognize this pattern to be consistent across a spectrum of transformer models designed for diverse tasks, such as LLMs, BERT, and Vision Transformer models. To tackle this, we propose per-channel activation quantization and show that these additional scaling factors can be reparameterized as exponential biases of weights, incurring a negligible cost. Our method, for the first time, can quantize both weights and activations in the LLaMA-13B to only 4-bit and achieves an average score of 63.1 on the common sense zero-shot reasoning tasks, which is only 5.8 lower than the full-precision model, significantly outperforming the previous state-of-the-art by 12.7 points. Code is available at: https://github.com/nbasyl/LLM-FP4.