mlsanityData to 7 Oct 2026

Temporally-Resolved Token Attribution Reveals the Generation Dynamics of Diffusion Language Models

arXiv 2610.01177 ↗First posted 1 Oct 2026cs.CL cs.AI cs.LGcited by 0

On the map: Diffusion models → Discrete and text diffusion

Abstract (arXiv)

This work presents Diffusion Layer Integrated Gradients (DLIG), a token attribution method for diffusion language models (DLMs) that extends Integrated Gradients (IG~citesundararajan2017axiomatic) 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.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Discrete and text diffusion, one of 14 directions in Diffusion models. Builds on 36 of its 46 references that are in the map. First posted on arXiv 1 Oct 2026.

Builds on

36 of 46 references.

Closest ancestors · 8

Same direction, newest first

  1. 2026TimeROME-DLM: Temporal Causal Tracing and Low-Rank Inference-Time Knowledge Editing for Masked Diffusion Language Models11 Jun
  2. 2026DLM-SWAI: Steering Diffusion Language Models Before They Unmask28 May
  3. 2026Where Should Diffusion Enter a Language Model? Geometry-Guided Hidden-State Replacement14 May
  4. 2025Activation Steering for Masked Diffusion Language Models30 Dec
  5. 2025Dream 7B: Diffusion Large Language Models21 Aug
  6. 2025Seed Diffusion: A Large-Scale Diffusion Language Model with High-Speed Inference4 Aug
  7. 2025Continuous Diffusion Model for Language Modeling17 Feb
  8. 2024Scaling Diffusion Language Models via Adaptation from Autoregressive Models23 Oct

Other references · 28

  1. 2020Denoising Diffusion Probabilistic Modelscited by 20,851
  2. 2015Deep Unsupervised Learning using Nonequilibrium Thermodynamicscited by 7,005
  3. 2016"Why Should I Trust You?": Explaining the Predictions of Any Classifiercited by 5,040
  4. 2017Axiomatic Attribution for Deep Networkscited by 3,350
  5. 2022Locating and Editing Factual Associations in GPTcited by 2,609
  6. 2016Understanding intermediate layers using linear classifier probescited by 1,279
  7. 2022Diffusion-LM Improves Controllable Text Generationcited by 1,006
  8. 2024Training Large Language Models to Reason in a Continuous Latent Spacecited by 763
  9. 2024Mechanistic Interpretability for AI Safety -- A Reviewcited by 415
  10. 2016A Corpus and Evaluation Framework for Deeper Understanding of Commonsense Storiescited by 413
  11. 2018WiC: the Word-in-Context Dataset for Evaluating Context-Sensitive Meaning Representationscited by 370
  12. 2023Towards Best Practices of Activation Patching in Language Models: Metrics and Methodscited by 328
16 more
  1. 2019A Multiscale Visualization of Attention in the Transformer Modelcited by 292
  2. 2024The pitfalls of next-token predictioncited by 156
  3. 2025A Survey on Sparse Autoencoders: Interpreting the Internal Mechanisms of Large Language Modelscited by 73
  4. 2021Discretized Integrated Gradients for Explaining Language Modelscited by 45
  5. 2023Using Captum to Explain Generative Language Modelscited by 33
  6. 2026Do Sparse Autoencoders Capture Concept Manifolds?cited by 25
  7. 2026Manifold Steering Reveals the Shared Geometry of Neural Network Representation and Behaviorcited by 22
  8. 2025DiffuGuard: How Intrinsic Safety is Lost and Found in Diffusion Large Language Modelscited by 11
  9. 2026FLARE: Diffusion for Hybrid Language Modelcited by 4
  10. 2026Observable Patterns Are Not Explanations: A Causal-Geometric Analysis of Latent Reasoning Modelscited by 3
  11. 2026Structuring Sparsity: Block-Sparse Featurizers Capture Visual Concept Manifoldscited by 3
  12. 2025Phonetic Perturbations Reveal Tokenizer-Rooted Safety Gaps in LLMscited by 2
  13. 2025ConceptGuard: Neuro-Symbolic Safety Guardrails via Sparse Interpretable Jailbreak Conceptscited by 2
  14. 2026Diffusion-CAM: Faithful Visual Explanations for dMLLMscited by 1
  15. 2026Diffusion Integrated Gradients: Controllable Path Generation for Flexible Feature Attributioncited by 1
  16. 2026When Do Concepts Become Functionally Sufficient During Language-Model Training?cited by 1

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Just on Time: Token-Level Early Stopping for Diffusion Language Models
  2. 2026Measuring Temporal Linguistic Emergence in Diffusion Language Models
  3. 2025On the Role of Discreteness in Diffusion LLMs
  4. 2026Hierarchical Continuous Diffusion Language Models
  5. 2026Diffusion Language Models: An Experimental Analysis
  6. 2026Time-Annealed Perturbation Sampling: Diverse Generation for Diffusion Language Models
  7. 2026Window-Diffusion: Accelerating Diffusion Language Model Inference with Windowed Token Pruning and Caching
  8. 2026DA-DLM: Explicitly Modeling Token Dependencies in Diffusion Language Models