Research
Temporally-Resolved Token Attribution Reveals the Generation Dynamics of Diffusion Language Models
arXiv:2610.01177v1 Announce Type: cross Abstract: This work presents Diffusion Layer Integrated Gradients (DLIG), a token attribution method for diffusion language models (DLMs) that extends Integrate
arXiv:2610.01177v1 Announce Type: cross Abstract: This work presents Diffusion Layer Integrated Gradients (DLIG), a token attribution method for diffusion language models (DLMs) that extends Integrated Gradients (IG~ite{sundararajan2017axiomatic}) to arbitrary layers and denoising steps. DLIG attributes a DLM's progressive commitment to a self-generated or fixed completion for an input prompt. We establish direct correspondences between DLIG and the IG axioms of completeness, implementation invariance, linearity, and symmetry preservation. As a lightweight complement to interventional analysis, DLIG provides an inexpensive first check of mechanistic hypotheses across the denoising trajectory. We demonstrate this on word-sense disambiguation, multi-hop graph reasoning, and sentence infilling, revealing how DLMs draw on inputs across positions, layers, and denoising steps.
Related
- Unified Energy for Invariant and Independent Decoding in Diffusion Language Models
- Parallelism and Generation Order in Masked Diffusion Language Models: Limits Today, Potential Tomorrow
- TAD: Temporal-Aware Trajectory Self-Distillation for Fast and Accurate Diffusion LLM
Source: arXiv cs.AI | 2026-10-02