DLM-One: Diffusion Language Models for One-Step Sequence Generation
cs.CL, cs.LG, stat.ML
Submitted: 2025-05-30
Updated: 2026-09-02
Code: https://github.com/Shark-NLP/DiffuSeq
License: http://creativecommons.org/licenses/by/4.0/
The gist: This paper introduces DLM-One, a score-distillation-based framework for one-step sequence generation with continuous diffusion language models (DLMs).
Terminology
Abstract
This paper introduces DLM-One, a score-distillation-based framework for one-step sequence generation with continuous diffusion language models (DLMs). DLM-One eliminates iterative refinement by aligning the scores of a student model's outputs with the score function of a pretrained teacher DLM in the forward-diffused noisy space. We demonstrate that our framework is architecture-agnostic and robust across diverse continuous manifolds, including standard token embedding spaces and logit simplex spaces. Through experiments on multiple representative DLMs, we show that DLM-One achieves up to about 2000 times speedup in sampling steps and about 500 times in wall-clock time, while maintaining competitive performance on benchmark text generation tasks. We further analyze failure modes in language-domain diffusion distillation and propose an adversarially-regularized two-stage training scheme to prevent student degeneration. Our findings position one-step score distillation as a viable path for the efficient deployment of continuous diffusion models operating in continuous space for natural language processing.
Sources
- GPT-4 Technical Report
- Structured Denoising Diffusion Models in Discrete State-Spaces
- Qwen Technical Report
- PaLM: Scaling Language Modeling with Pathways
- Quasar: Datasets for Question Answering by Search and Reading
- Continuous diffusion for categorical data
- Consistency Models Made Easy
- DiffuSeq: Sequence to Sequence Text Generation with Diffusion Models
- Scaling Diffusion Language Models via Adaptation from Autoregressive Models
- The Llama 3 Herd of Models
- Plug-and-Play Controllable Generation for Discrete Masked Models
- Classifier-Free Diffusion Guidance
- Neural CRF Model for Sentence Alignment in Text Simplification
- Challenges and Applications of Large Language Models
- CTRL: A Conditional Transformer Language Model for Controllable Generation
- A Comprehensive Survey of Accelerated Generation Techniques in Large Language Models
- Discrete Copula Diffusion
- Improved Denoising Diffusion Probabilistic Models
- Large Language Diffusion Models
- Neural Discrete Representation Learning
Related papers
- Exploring Solution Divergence and Its Effect on Large Language Model Problem Solving
- Ishigaki-IDS-Bench: A Benchmark for Generating Information Delivery Specification from BIM Information Requirements
- Subliminal Steering: Stronger Encoding of Hidden Signals
- MedStruct-S: A Benchmark for Key Discovery, Key-Conditioned QA and Semi-Structured Extraction from OCR Clinical Reports
- The End of Transformers? On Challenging Attention and the Rise of Sub-Quadratic Architectures
- Untangling the Mechanisms of Misleading Context in Medical Question Answering