Beyond Model Weights: Sampling-Dependent Fairness in Diffusion Language Models
Diffusion language models (DLMs) are emerging as a practical alternative to autoregressive language models (ARLMs), yet their fairness remains largely uncharacterized.
ProofPaper ↗
Key points
- DLMs introduce sampling controls with no direct autoregressive counterpart, such as block length and remasking strategy, that directly set their speed--quality operating point.
- This raises a basic question: does a DLM's measured social bias remain stable across these settings?
- Across LLaDA, LLaDA-1.5, and Dream on Open-BBQ, we find that all three models exhibit social bias, but both the magnitude of measured bias and the relative ordering of models by bias are highly sensitive to the sampling configuration.
- Our results show that fairness in DLMs cannot be characterized independently of the sampling configuration, even after debiasing.
Sources (1)
- [1]Beyond Model Weights: Sampling-Dependent Fairness in Diffusion Language ModelsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 20, 03:49 AM
Diffusion language models (DLMs) are emerging as a practical alternative to autoregressive language models (ARLMs), yet their fairness remains largely uncharacterized.
DLMs introduce sampling controls with no direct autoregressive counterpart, such as block length and remasking strategy, that directly set their speed--quality operating point.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 20, 2026TeCoP: Accelerating Diffusion Large Language Model Decoding via Tentative Context Pipelining
- Oct 20, 2026Large Flow Language Models
- Oct 19, 2026Balance-Aware Pivot Selection for Parallel Diffusion LLM Decoding
- Oct 19, 2026Keep the Lineages Alive: Diversity-Preserving SMC for Diffusion LLM Decoding
- Oct 18, 2026Enhanced Masked Diffusion Unlearning
- Oct 17, 2026Who Gets Unmasked? Sampler-Induced Bias in Diffusion Language Models