Data to 7 Oct 2026

DiffuCoder: Understanding and Improving Masked Diffusion Models for Code Generation

arXiv 2506.20639 ↗First posted 25 Jun 2025cs.CLcited by 184

On the map: Diffusion models → Discrete and text diffusion

Abstract (arXiv)

Diffusion large language models (dLLMs) are compelling alternatives to autoregressive (AR) models because their denoising models operate over the entire sequence. The global planning and iterative refinement features of dLLMs are particularly useful for code generation. However, current training and inference mechanisms for dLLMs in coding are still under-explored. To demystify the decoding behavior of dLLMs and unlock their potential for coding, we systematically investigate their denoising processes and reinforcement learning (RL) methods. We train a 7B dLLM, DiffuCoder, on 130B tokens of code. Using this model as a testbed, we analyze its decoding behavior, revealing how it differs from that of AR models: (1) dLLMs can decide how causal their generation should be without relying on semi-AR decoding, and (2) increasing the sampling temperature diversifies not only token choices but also their generation order. This diversity creates a rich search space for RL rollouts. For RL training, to reduce the variance of token log-likelihood estimates and maintain training efficiency, we propose coupled-GRPO, a novel sampling scheme that constructs complementary mask noise for completions used in training. In our experiments, coupled-GRPO significantly improves DiffuCoder's performance on code generation benchmarks (+4.4% on EvalPlus) and reduces reliance on AR bias during decoding. Our work provides deeper insight into the machinery of dLLM generation and offers an effective, diffusion-native RL training framework. https://github.com/apple/ml-diffucoder.

Key facts. Cited by 184 papers in the map (data through 7 Oct 2026). Placed in Discrete and text diffusion, one of 14 directions in Diffusion models. Builds on 68 of its 80 references that are in the map. First posted on arXiv 25 Jun 2025.

Builds on

68 of 80 references.

Closest ancestors · 20

Same direction, newest first

8 more

Other references · 48

2020Denoising Diffusion Probabilistic Modelscited by 20,851
2023GPT-4 Technical Reportcited by 18,132
2017Proximal Policy Optimization Algorithmscited by 14,441
2023LLaMA: Open and Efficient Foundation Language Modelscited by 13,958
2021Training Verifiers to Solve Math Word Problemscited by 9,743
2020Score-Based Generative Modeling through Stochastic Differential Equationscited by 8,964
2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 8,851
2021Evaluating Large Language Models Trained on Codecited by 8,053
2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 7,916
2021Program Synthesis with Large Language Modelscited by 3,509
2023Efficient Streaming Language Models with Attention Sinkscited by 2,115
2023Code Llama: Open Foundation Models for Codecited by 2,105

36 more

2021Structured Denoising Diffusion Models in Discrete State-Spacescited by 1,665
2023Is Your Code Generated by ChatGPT Really Correct? Rigorous Evaluation of Large Language Models for Code Generationcited by 1,332
2019ZeRO: Memory Optimizations Toward Training Trillion Parameter Modelscited by 1,278
2024Qwen2.5-Coder Technical Reportcited by 1,151
2024The FineWeb Datasets: Decanting the Web for the Finest Text Data at Scalecited by 1,104
2022MaskGIT: Masked Generative Image Transformercited by 1,082
2022Diffusion-LM Improves Controllable Text Generationcited by 1,006
2025Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?cited by 948
2023Training Diffusion Models with Reinforcement Learningcited by 939
2023PyTorch FSDP: Experiences on Scaling Fully Sharded Data Parallelcited by 641
2021Argmax Flows and Multinomial Diffusion: Learning Categorical Distributionscited by 566
2024BigCodeBench: Benchmarking Code Generation with Diverse Function Calls and Complex Instructionscited by 527
2024StarCoder 2 and The Stack v2: The Next Generationcited by 484
2025The Entropy Mechanism of Reinforcement Learning for Reasoning Language Modelscited by 463
2025Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decodingcited by 404
2022Analog Bits: Generating Discrete Data using Diffusion Models with Self-Conditioningcited by 362
2025LLaDA 1.5: Variance-Reduced Preference Optimization for Large Language Diffusion Modelscited by 273
2024Diffusion Policy Policy Optimizationcited by 255
2025MMaDA: Multimodal Large Diffusion Language Modelscited by 241
2025Mercury: Ultra-Fast Language Models Based on Diffusioncited by 225
2024When Attention Sink Emerges in Language Models: An Empirical Viewcited by 200
2025Spurious Rewards: Rethinking Training Signals in RLVRcited by 199
2025ProRL: Prolonged Reinforcement Learning Expands Reasoning Boundaries in Large Language Modelscited by 187
2025d1: Scaling Reasoning in Diffusion Large Language Models via Reinforcement Learningcited by 178
2025dLLM-Cache: Accelerating Diffusion Large Language Models with Adaptive Cachingcited by 168
2025The Unreasonable Effectiveness of Entropy Minimization in LLM Reasoningcited by 160
2025Llama-Nemotron: Efficient Reasoning Modelscited by 151
2024OpenCoder: The Open Cookbook for Top-Tier Code Large Language Modelscited by 127
2025ACECODER: Acing Coder RL via Automated Test-Case Synthesiscited by 89
2025LaViDa: A Large Diffusion Language Model for Multimodal Understandingcited by 81
2025AceReason-Nemotron 1.1: Advancing Math and Code Reasoning through SFT and RL Synergycited by 75
2023Lemur: Harmonizing Natural Language and Code for Language Agentscited by 71
2025Teaching Language Models to Critique via Reinforcement Learningcited by 38
2024Programming Every Example: Lifting Pre-training Data Quality Like Experts at Scalecited by 37
2025Fine-Tuning Discrete Diffusion Models with Policy Gradient Methodscited by 36
2025Unifying Autoregressive and Diffusion-Based Sequence Generationcited by 17

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 78; 2026: 106.

184 papers.

Sort:

By area:

2025A Survey of Reinforcement Learning for Large Reasoning Modelscited by 175
2025SDAR: A Synergistic Diffusion-AutoRegression Paradigm for Scalable Sequence Generationcited by 118
2025Dream-Coder 7B: An Open Diffusion Language Model for Codecited by 81
2025A Survey on Diffusion Language Modelscited by 75
2025dParallel: Learnable Parallel Decoding for dLLMscited by 59
2025LLaDA-MoE: A Sparse MoE Diffusion Language Modelcited by 54
2025Any-Order Flexible Length Masked Diffusioncited by 48
2025Your LLM Knows the Future: Uncovering Its Multi-Token Prediction Potentialcited by 46
2025Diffusion Language Models are Super Data Learnerscited by 45
2025ParallelBench: Understanding the Trade-offs of Parallel Decoding in Diffusion LLMscited by 42
2025Fine-Tuning Masked Diffusion for Provable Self-Correctioncited by 39
2025DPad: Efficient Diffusion Language Models with Suffix Dropoutcited by 38

38 more

2025Time Is a Feature: Exploiting Temporal Dynamics in Diffusion Language Modelscited by 33
2025The Devil behind the mask: An emergent safety vulnerability of Diffusion LLMscited by 32
2025Continuously Augmented Discrete Diffusion model for Categorical Generative Modelingcited by 32
2025AdaBlock-dLLM: Semantic-Aware Diffusion LLM Inference via Adaptive Block Sizecited by 30
2025Improving Reasoning for Diffusion Language Models via Group Diffusion Policy Optimizationcited by 30
2025Principled RL for Diffusion LLMs Emerges from a Sequence-Level Perspectivecited by 30
2025Efficient-DLM: From Autoregressive to Diffusion Language Models, and Beyond in Speedcited by 30
2025Learning Unmasking Policies for Diffusion Language Modelscited by 29
2025Coevolutionary Continuous Discrete Diffusion: Make Your Diffusion Language Model a Latent Reasonercited by 28
2025A Survey on Parallel Text Generation: From Parallel Decoding to Diffusion Language Modelscited by 27
2025d2: Improving Reasoning in Diffusion Language Models via Trajectory Likelihood Estimationcited by 27
2026ELF: Embedded Language Flowscited by 27
2025SpecDiff-2: Scaling Diffusion Drafter Alignment For Faster Speculative Decodingcited by 26
2025Set Block Decoding is a Language Model Inference Acceleratorcited by 22
2025Diffusion LLM with Native Variable Generation Lengths: Let [EOS] Lead the Waycited by 22
2025From Next-Token to Next-Block: A Principled Adaptation Path for Diffusion LLMscited by 22
2025Structuring The Future: Diffusion LLM Speculative Decoding via Calibrated Draft Graphscited by 21
2025Training Optimal Large Diffusion Language Modelscited by 21
2025CDLM: Consistency Diffusion Language Models For Faster Samplingcited by 20
2026Nemotron-Labs-Diffusion: A Tri-Mode Language Model Unifying Autoregressive, Diffusion, and Self-Speculation Decodingcited by 19
2025Dream-VL & Dream-VLA: Open Vision-Language and Vision-Language-Action Models with Diffusion Language Model Backbonecited by 18
2025DiFFPO: Training Diffusion LLMs to Reason Fast and Furious via Reinforcement Learningcited by 17
2025LoPA: Scaling dLLM Inference via Lookahead Parallel Decodingcited by 17
2025Inpainting-Guided Policy Optimization for Diffusion Large Language Modelscited by 16
2025Exploring the Potential of Diffusion Large Language Models in Code Generationcited by 16
2025FS-DFM: Fast and Accurate Long Text Generation with Few-Step Diffusion Language Modelscited by 16
2026DreamOn: Diffusion Language Models For Code Infilling Beyond Fixed-size Canvascited by 16
2025Enhancing Reasoning for Diffusion LLMs via Distribution Matching Policy Optimizationcited by 15
2025Soft-Masked Diffusion Language Modelscited by 15
2025Lookahead Unmasking Elicits Accurate Decoding in Diffusion Language Modelscited by 15
2025d-TreeRPO: Towards More Reliable Policy Optimization for Diffusion Language Modelscited by 14
2026DMax: Aggressive Parallel Decoding for dLLMscited by 14
2025Constrained Decoding of Diffusion LLMs with Context-Free Grammarscited by 13
2025TR2-D2: Tree Search Guided Trajectory-Aware Fine-Tuning for Discrete Diffusioncited by 13
2025Saber: An Efficient Sampling with Adaptive Acceleration and Backtracking Enhanced Remasking for Diffusion Language Modelcited by 13
2025Parallel Sampling from Masked Diffusion Models via Conditional Independence Testingcited by 13
2025From Code Foundation Models to Agents and Applications: A Comprehensive Survey and Practical Guide to Code Intelligencecited by 13
2026Stable-DiffCoder: Pushing the Frontier of Code Diffusion Large Language Modelcited by 12

50 most cited of 184 All 184 (JSON)

Similar, not cited

Close in wording, no citation either way