SoloQ: Calibration-Free Quantization for Diffusion Language Models

2026-10-05

AI Summary

This study addresses the challenge that quantizing diffusion language models typically requires calibration data, as activation distributions vary drastically during denoising. To overcome this, we propose a calibration-free quantization framework that introduces a structured K-RPBH rotation combined with a lightweight rescaling mechanism. This approach maps both weights and activations onto a predictable normalized rotational basis, enabling data-independent low-bit inference via distribution-matching codebooks. Furthermore, the framework supports hardware-native NVFP4 formats and KV cache compression for block diffusion models. Experimental results demonstrate that under 4-bit quantization, our method surpasses calibration-dependent baselines in accuracy while reducing peak memory consumption by 2.61× and accelerating end-to-end inference speed by 2.24×.

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. Low-bit weight-activation quantization is therefore attractive, yet existing dLLM post-training quantization methods rely on calibration data despite activation distributions shifting across masking states and denoising steps. We present SoloQ, a calibration-free quantization framework that maps weights and activations into a normalized rotated basis with a predictable marginal distribution, enabling data-independent quantization. SoloQ combines a structured K-RPBH rotation with a lightweight rescaling correction for calibration-free quantization. Its predictable post-rotation distribution supports both distribution-matched codebooks and hardware-native NVFP4. For block-diffusion models, SoloQ further applies commit-time KV-cache quantization to compress persistent states without perturbing the actively denoised block. Across full-sequence dLLMs (LLaDA and Dream) and block-diffusion dLLMs(Fast-dLLM v2 and Nemotron-Labs-Diffusion), SoloQ retains accuracy under 4-bit quantization and outperforms calibration-based baselines on knowledge- and reasoning-intensive benchmarks. With NVFP4, SoloQ reduces peak memory by up to 2.61X and accelerates end-to-end inference by up to 2.24X.

Problem

• Diffusion Language Models
• Quantization
• Calibration-Free
• Inference Efficiency
• KV-cache

Innovation

• Calibration-Free Quantization
• Diffusion Language Models
• K-RPBH Rotation
• Commit-time KV-cache Quantization
• NVFP4