Bit-by-Bit: Progressive QAT Strategy with Outlier Channel Splitting for Stable Low-Bit LLMs

TL;DR

This paper introduces Bit-by-Bit, a progressive quantization-aware training framework with outlier channel splitting, enabling stable low-bit LLM training, multi-bit deployment, and efficient custom kernels, significantly improving performance and speed.

Contribution

The paper proposes a novel progressive QAT method with outlier channel splitting, supporting multi-bit deployment and custom kernels for stable, efficient ultra-low precision LLM training.

Findings

Outperforms baselines like BitDistiller and EfficientQAT on Llama models.

Achieves up to 11× speedup with custom 2-bit kernels.

Maintains low perplexity loss with only 2.25 PPL increase on WikiText2.

Abstract

Training LLMs at ultra-low precision remains a formidable challenge. Direct low-bit QAT often suffers from convergence instability and substantial training costs, exacerbated by quantization noise from heavy-tailed outlier channels and error accumulation across layers. To address these issues, we present Bit-by-Bit, a progressive QAT framework with outlier channel splitting. Our approach integrates three key components: (1) block-wise progressive training that reduces precision stage by stage, ensuring stable initialization for low-bit optimization; (2) nested structure of integer quantization grids to enable a "train once, deploy any precision" paradigm, allowing a single model to support multiple bit-widths without retraining; (3) rounding-aware outlier channel splitting, which mitigates quantization error while acting as an identity transform that preserves the quantized outputs. Furthermore, we follow microscaling groups with E4M3 scales, capturing dynamic activation ranges in alignment with OCP/NVIDIA standards. To address the lack of efficient 2-bit kernels, we developed custom operators for both W2A2 and W2A16 configurations, achieving up to 11$\times$ speedup over BF16. Under W2A2 settings, Bit-by-Bit significantly outperforms baselines like BitDistiller and EfficientQAT on both Llama2/3, achieving a loss of only 2.25 WikiText2 PPL compared to full-precision models.