HiFloat4 Format for Language Model Pre-training on Ascend NPUs

TL;DR

This paper evaluates the HiFloat4 FP4 format for efficient large-scale language model training on Huawei Ascend NPUs, comparing it with other FP4 formats and exploring stabilization techniques.

Contribution

It provides a comprehensive empirical comparison of HiFloat4 with MXFP4 for large-scale training on NPUs, including stabilization methods for FP4.

Findings

HiFloat4 achieves comparable accuracy to full-precision training with stabilization.

FP4 formats significantly improve compute throughput and memory efficiency.

Stabilization techniques reduce numerical errors to within 1% of full-precision baselines.

Abstract

Large foundation models have become central to modern machine learning, with performance scaling predictably with model size and data. However, training and deploying such models incur substantial computational and memory costs, motivating the development of low-precision training techniques. Recent work has demonstrated that 4-bit floating-point (FP4) formats--such as MXFP4 and NVFP4--can be successfully applied to linear GEMM operations in large language models (LLMs), achieving up to 4x improvements in compute throughput and memory efficiency compared to higher-precision baselines. In this work, we investigate the recently proposed HiFloat4 FP4 format for Huawei Ascend NPUs and systematically compare it with MXFP4 in large-scale training settings. All experiments are conducted on Ascend NPU clusters, with linear and expert GEMM operations performed entirely in FP4 precision. We evaluate both dense architectures (e.g., Pangu and LLaMA-style models) and mixture-of-experts (MoE) models, where both standard linear layers and expert-specific GEMMs operate in FP4. Furthermore, we explore stabilization techniques tailored to FP4 training that significantly reduce numerical degradation, maintaining relative error within 1% of full-precision baselines while preserving the efficiency benefits of 4-bit computation. Our results provide a comprehensive empirical study of FP4 training on NPUs and highlight the practical trade-offs between FP4 formats in large-scale dense and MoE models.