让微调模型随噪声变化自适应，提升扩散语言模型生成效果
NaRA: Noise-Aware LoRA for Parameter-Efficient Fine-Tuning of Diffusion LLMs
- 用噪声水平控制低秩更新矩阵，动态适应扩散过程
- 在常识推理等任务上显著优于传统LoRA方法
- 适合关注扩散模型高效微调的研究者和开发者
扩散大语言模型（dLLMs）作为一种有前景的非自回归生成范式受到关注。由于全量微调计算成本过高，参数高效微调（PEFT）已成为标准方法。然而，现有PEFT方法（如LoRA）最初为自回归模型设计，依赖与噪声水平无关的静态参数，忽略了扩散过程中输入分布和生成难度随去噪轨迹变化的内在动态特性，因此对dLLMs表现不佳。为此，我们提出噪声感知低秩适配（NaRA），引入由轻量级全局共享超网络根据噪声水平生成的低秩核心矩阵。该设计使更新矩阵能沿扩散过程连续变化，同时保持参数和延迟开销可忽略。我们为NaRA框架提供理论支持，并在常识推理、数学推理和代码生成基准上实证证明其持续优于噪声无关基线。代码已公开于https://github.com/generaldi/NaRA。
原文摘要 · Abstract (English)
Diffusion Large Language Models (dLLMs) have emerged as a promising non-autoregressive generative paradigm. Given the prohibitive computational cost of full fine-tuning, Parameter-Efficient Fine-Tuning (PEFT) has become the standard approach. However, existing PEFT methods (e.g., LoRA), originally tailored for autoregressive models, rely on static parameters that are agnostic to the noise level. Consequently, they ignore the intrinsic dynamics of the diffusion process, where input distributions and generation difficulty shift significantly along the denoising trajectory, rendering them suboptimal for dLLMs. To address this, we propose Noise-aware Low-Rank Adaptation (NaRA), which introduces a low-rank core matrix generated by a lightweight, globally shared hypernetwork conditioned on the noise level. This design enables the update matrices to vary continuously along the diffusion process while keeping parameter and latency overhead negligible. We provide a theoretical justification for the proposed NaRA framework and empirically demonstrate consistent improvements over noise-agnostic baselines across commonsense reasoning, mathematical reasoning, and code generation benchmarks. Our code is available at https://github.com/generaldi/NaRA.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。