BitNet v2: Native 4-bit Activations with Hadamard Transformation for 1-bit LLMs
This addresses the efficiency bottleneck for deploying 1-bit LLMs, though it appears incremental over previous BitNet work.
The paper tackles the problem of activation outliers hindering efficient deployment of 1-bit LLMs by introducing BitNet v2, which enables native 4-bit activation quantization using an online Hadamard transformation. The result is minimal performance degradation with significantly reduced memory and computational costs for batched inference.
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.