SoloQ：面向扩散语言模型的无校准量化
原标题：Paper page - SoloQ: Calibration-Free Quantization for Diffusion Language Models
AI 摘要
该论文提出 SoloQ，一种面向扩散语言模型（dLLM）的无校准量化框架。现有 dLLM 训练后量化方法依赖校准数据，但激活分布会随掩码状态和去噪步骤变化。SoloQ 通过将权重和激活映射到归一化旋转基，实现数据无关的量化，并结合结构化 K-RPBH 旋转与轻量重缩放校正，支持分布匹配码本和硬件原生 NVFP4。在 LLaDA、Dream、Fast-dLLM v2 和 Nemotron-Labs-Diffusion 上，SoloQ 在 4-bit 量化下保持精度并优于基于校准的基线，NVFP4 下峰值内存最多降低 2.61 倍，端到端推理最多加速 2.24 倍。
正文节选
SoloQ: Calibration-Free Quantization for Diffusion Language Models Abstract Diffusion large language models dLLMs) have emerged as a promising alternative to autoregressive language models through bidirectional diffusion-based token generation. However, their growing model sizes and high inference costs make efficient deployment challenging: full-sequence denoising repeatedly invokes compute-intensive forward passes, while block-diffusion models additionally introduce a memory-intensive KV-cache