TAD: Temporal-Aware Trajectory Self-Distillation for Fast and Accurate Diffusion LLM

• URL: http://arxiv.org/abs/2605.09536v1
• Date: Sun, 10 May 2026 13:38:53 GMT
• Title: TAD: Temporal-Aware Trajectory Self-Distillation for Fast and Accurate Diffusion LLM
• Abstract summary: Diffusion large language models (dLLMs) offer a promising paradigm for parallel text generation.<n>They face an accuracy-parallelism trade-off, where increasing tokens per forward (TPF) often degrades generation quality.<n>Existing acceleration methods often gain speed at the cost of accuracy.<n>We propose TAD, a Temporal-Aware trajectory self-Distillation framework.
• Score: 18.69783590125584
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Diffusion large language models (dLLMs) offer a promising paradigm for parallel text generation, but in practice they face an accuracy-parallelism trade-off, where increasing tokens per forward (TPF) often degrades generation quality. Existing acceleration methods often gain speed at the cost of accuracy. To address this limitation, we propose TAD, a Temporal-Aware trajectory self-Distillation framework. During data construction, we condition a teacher model on both the prompt and the ground-truth response to generate decoding trajectories, recording the intermediate masked states throughout the process. Based on how many decoding steps remain before each masked token is revealed, we partition masked positions into near and distant subsets. For near tokens, we train the student with a hard cross-entropy loss using the teacher trajectory tokens as labels, encouraging confident predictions for tokens that are about to be decoded. For distant tokens, we apply a soft KL divergence loss between the teacher and student token distributions, providing softer supervision and preserving future planning knowledge. This temporal-aware partition naturally gives rise to two deployment configurations: a Quality model that prioritizes accuracy and a Speed model that favors more aggressive acceleration. Experiments show that TAD consistently improves the accuracy-parallelism trade-off. On LLaDA, it raises average accuracy from 46.2\% to 51.6\% with the Quality model and average AUP from 46.2 to 257.1 with the Speed model. Our code is available at: https://github.com/BHmingyang/TAD

Related papers

• Temporal Self-Distillation: Faster Inference in Discrete Diffusion Language Models [11.929379448468568]
  Diffusion language models (dLLMs) promise fast inference by generating multiple tokens in parallel, but suffer severe performance degradation when parallel decoding is pushed too aggressively.<n>We introduce Temporal Self-Distillation (TSD), a simple on-policy method that trains dLLMs for fast inference by distilling predictions across time.
  arXiv  Detail & Related papers  (2026-09-14T07:59:45Z)
• TACG: Trajectory-Aware Commit Gating for Diffusion Language Model Decoding [38.72024551312125]
  Trajectory-Aware Commit Gating (TACG) is a gate-level decoder that anchors token identities to the base posterior.<n>We evaluate TACG on LLaDA, Dream, and LLaDA2-Mini benchmarks.
  arXiv  Detail & Related papers  (2026-07-03T11:45:21Z)
• Roll Out and Roll Back: Diffusion LLMs are Their Own Efficiency Teachers [76.15132587294862]
  Wide-In, Narrow-Out (WINO) is a training-free decoding algorithm that enables revokable parallel generation.<n>WINO+ injects the verified denoising trajectories produced by WINO into model parameters, aligning training with efficient inference.<n>Experiments on LLaDA and MMaDA show that WINO improves both quality and efficiency, while WINO+ further strengthens this progression.
  arXiv  Detail & Related papers  (2026-05-16T11:27:40Z)
• d3LLM: Ultra-Fast Diffusion LLM using Pseudo-Trajectory Distillation [31.922313594074925]
  Diffusion large language models (dLLMs) offer capabilities beyond those of autoregressive (AR) LLMs.<n>Current methods typically focus on only one-side of the coin, targeting either efficiency or performance.<n>We propose d3LLM (Pseudo-Distilled Diffusion Large Language Model), striking a balance between accuracy and parallelism.
  arXiv  Detail & Related papers  (2026-01-12T14:25:36Z)
• Teach Diffusion Language Models to Learn from Their Own Mistakes [45.68746718883178]
  Masked Diffusion Language Models (DLMs) achieve significant speed by generating multiple tokens in parallel.<n> parallel sampling approach will introduce strong dependency errors and cause quality to deteriorate rapidly as the generation step size grows.<n>We propose Decoupled Self-Correction to maintain high-quality multi-token generation.
  arXiv  Detail & Related papers  (2026-01-10T05:04:33Z)
• CD4LM: Consistency Distillation and aDaptive Decoding for Diffusion Language Models [27.070045950001532]
  CD4LM is a framework that decouples training from inference.<n>On GSM8K, CD4LM matches the LLaDA baseline with a 5.18x wall-clock speedup.
  arXiv  Detail & Related papers  (2026-01-05T16:09:22Z)
• Pipeline Parallelism is All You Need for Optimized Early-Exit Based Self-Speculative Decoding [73.67253077506672]
  Large language models (LLMs) deliver impressive generation quality, but incur very high inference cost.<n>Early-exit based self-speculative decoding (EESD) has emerged to mitigate this cost.<n>We propose Pipeline-Parallel Self-Speculative Decoding (PPSD) that fully pipelines the draft and verification work.
  arXiv  Detail & Related papers  (2025-09-19T04:51:41Z)
• Diffusion Language Models Know the Answer Before Decoding [56.96815863705218]
  Diffusion language models (DLMs) have emerged as an alternative to autoregressive approaches.<n>Our work highlights and leverage an overlooked property of DLMs early answer convergence.<n>We introduce Prophet, a training-free fast decoding paradigm that enables early commit decoding.
  arXiv  Detail & Related papers  (2025-08-27T15:40:25Z)
• MDPO: Overcoming the Training-Inference Divide of Masked Diffusion Language Models [28.79185891706149]
  Diffusion language models suffer from a key discrepancy between training and inference.<n>We propose a novel Masked Diffusion Policy Optimization (MDPO) to exploit the Markov property diffusion.<n>Our findings establish great potential for investigating the discrepancy between pre-training and inference of MDLMs.
  arXiv  Detail & Related papers  (2025-08-18T17:58:13Z)
• REPA Works Until It Doesn't: Early-Stopped, Holistic Alignment Supercharges Diffusion Training [58.33728862521732]
  Diffusion Transformers (DiTs) deliver state-of-the-art image quality, yet their training remains notoriously slow.<n>A recent remedy -- representation alignment (REPA) that matches DiT hidden features to those of a non-generative teacher (e.g. DINO) -- dramatically accelerates the early epochs but plateaus or even degrades performance later.<n>We trace this failure to a capacity mismatch: once the generative student begins modelling the joint data distribution, the teacher's lower-dimensional embeddings and attention patterns become a straitjacket rather than a guide.<n>We introduce HASTE
  arXiv  Detail & Related papers  (2025-05-22T15:34:33Z)
• FIRP: Faster LLM inference via future intermediate representation prediction [54.897493351694195]
  FIRP generates multiple tokens instead of one at each decoding step. We conduct extensive experiments, showing a speedup ratio of 1.9x-3x in several models and datasets.
  arXiv  Detail & Related papers  (2024-10-27T15:53:49Z)
• Promises and Pitfalls of Generative Masked Language Modeling: Theoretical Framework and Practical Guidelines [74.42485647685272]
  We focus on Generative Masked Language Models (GMLMs) We train a model to fit conditional probabilities of the data distribution via masking, which are subsequently used as inputs to a Markov Chain to draw samples from the model. We adapt the T5 model for iteratively-refined parallel decoding, achieving 2-3x speedup in machine translation with minimal sacrifice in quality.
  arXiv  Detail & Related papers  (2024-07-22T18:00:00Z)
• T4P: Test-Time Training of Trajectory Prediction via Masked Autoencoder and Actor-specific Token Memory [39.021321011792786]
  Trajectory prediction is a challenging problem that requires considering interactions among multiple actors. Data-driven approaches have been used to address this complex problem, but they suffer from unreliable predictions under distribution shifts during test time. We propose several online learning methods using regression loss from the ground truth of observed data. Our method surpasses the performance of existing state-of-the-art online learning methods in terms of both prediction accuracy and computational efficiency.
  arXiv  Detail & Related papers  (2024-03-15T06:47:14Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.