OPTD: On-Policy Transition Distillation with Consistency-Guided Adaptive Compression for Few-Step Diffusion Language Models
On the map: Diffusion models → Discrete and text diffusion
Abstract (arXiv)
Diffusion language models (dLLMs) can predict many tokens in parallel, but accurate generation still requires many iterative denoising steps. Few-step distillation accelerates decoding by compressing multiple teacher steps into a single student transition. However, existing methods construct supervision on off-policy trajectories. At inference, the student's early parallel commitments alter the context of later predictions, so the states it actually visits drift away from the supervised ones--precisely when step compression is most aggressive. On-policy distillation is a natural remedy for this mismatch, but it leaves open how far each transition should advance: matching only the teacher's next action limits compression, while indiscriminately merging future actions can violate intermediate dependencies. To address this limitation, we propose OPTD, On-Policy Transition Distillation with consistency-guided adaptive compression. It samples partial states from the few-step student's own trajectories, uses a frozen, question-only teacher to identify outcome-aligned future candidates, and orders them by current-state confidence. The method then selects the longest prefix whose joint commitment preserves the teacher's rollout outcome. A set-bottleneck objective promotes every verified future candidate to the decoder's release threshold, while a frozen-teacher KL anchor regularizes all other active positions. Neither target construction nor training uses a gold response. Across four mathematical reasoning and code-generation benchmarks, OPTD consistently improves the quality--efficiency trade-off and attains the strongest overall quality-constrained AUP among the evaluated few-step baselines.
Key facts. Cited by 3 papers in the map (data through 14 Oct 2026). Placed in Discrete and text diffusion, one of 14 directions in Diffusion models. Builds on 12 of its 17 references that are in the map. First posted on arXiv 3 Aug 2026.
Builds on
12 of 17 references.
Closest ancestors · 2
Same direction, newest first
- 2025CDLM: Consistency Diffusion Language Models For Faster Sampling24 Nov
- 2025Diffusion LLMs Can Do Faster-Than-AR Inference via Discrete Diffusion Forcing8 Aug
Other references · 10
- 2021Training Verifiers to Solve Math Word Problemscited by 10,109
- 2021Evaluating Large Language Models Trained on Codecited by 8,320
- 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,371
- 2021Program Synthesis with Large Language Modelscited by 3,637
- 2021Argmax Flows and Multinomial Diffusion: Learning Categorical Distributionscited by 574
- 2025Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decodingcited by 420
- 2025dParallel: Learnable Parallel Decoding for dLLMscited by 65
- 2026d3LLM: Ultra-Fast Diffusion LLM using Pseudo-Trajectory Distillationcited by 37
- 2026Learning from the Self-future: On-policy Self-distillation for dLLMscited by 5
- 2026TAD: Temporal-Aware Trajectory Self-Distillation for Fast and Accurate Diffusion LLMcited by 2
Built on by
Citing papers per year: 2026: 3.
3 papers.
Similar, not cited
Close in wording, no citation either way
- 2026dOPSD: On-Policy Self-Distillation for Diffusion Language Models
- 2026Hesitation-Aware On-Policy Distillation for Diffusion Language Models
- 2026Temporal Self-Distillation: Faster Inference in Discrete Diffusion Language Models
- 2026Towards Unbiased On-Policy Distillation for Block Diffusion Language Models
- 2026Few-Step Diffusion Language Models via Trajectory Self-Distillation
- 2026JustOPD: On-Policy Distillation from Autoregressive to Diffusion Language Models
- 2026Data-Efficient Autoregressive-to-Diffusion Language Models via On-Policy Distillation
- 2026TRIMS: Trajectory-Ranked Instruction Masked Supervision for Diffusion Language Models