BitNet v2: Native 4-bit Activations with Hadamard Transformation for 1-bit LLMs
Research paper by Hongyu Wang and collaborators. Published 2025-04-25. Topics: Large Language Models, LLMS, BitNet, H-BitLinear, Hadamard transformation.
Resource information
- Authors
- Hongyu Wang, Shuming Ma, Furu Wei
- Research area
- Language technology
- Application domain
- General AI research
- AI methods
- Large language models
- Research topics
- cs.CL, cs.LG
- arXiv record date
- 2025-06-13
- Publication date
- 2025-04-25
- Publication
- arXiv
- Upstream last modified
- Jun 16, 2025
- Last record update
- Nov 3, 2026
Abstract
Efficient deployment of 1-bit Large Language Models (LLMs) is hindered by activation outliers, which complicate quantization to low bit-widths. We introduce BitNet v2, a novel framework enabling native 4-bit activation quantization for 1-bit LLMs. To tackle outliers in attention and feed-forward network activations, we propose H-BitLinear, a module applying an online Hadamard transformation prior to activation quantization. This transformation smooths sharp activation distributions into more Gaussian-like forms, suitable for low-bit representation. Experiments show BitNet v2 trained from scratch with 8-bit activations matches BitNet b1.58 performance. Crucially, BitNet v2 achieves minimal performance degradation when trained with native 4-bit activations, significantly reducing memory footprint and computational cost for batched inference.