# Diffusion-LLM-Papers We’ve collected a series of papers on Diffusion Language Models. Since our time is limited and we can't cover everything, **please feel free to submit a pull request to contribute.** ## Theoretical Basis **Deep Unsupervised Learning using Nonequilibrium Thermodynamics** 2015-3-12, [Paper](https://arxiv.org/abs/1503.03585) **Structured Denoising Diffusion Models in Discrete State-Spaces** 2021-7-7, [Paper](https://arxiv.org/abs/2107.03006) **Discrete Diffusion Modeling by Estimating the Ratios of the Data Distribution** 2023-10-25, [Paper](https://arxiv.org/abs/2310.16834) **Your Absorbing Discrete Diffusion Secretly Models the Conditional Distributions of Clean Data** 2024-6-6, [Paper](https://arxiv.org/abs/2406.03736) **Simplified and Generalized Masked Diffusion for Discrete Data** 2024-6-6, [Paper](https://arxiv.org/abs/2406.04329) **Simple and Effective Masked Diffusion Language Models** 2024-6-11, [Paper](https://arxiv.org/abs/2406.07524) ## Foundation Model **LLaDA: Large Language Diffusion Models** 2025-2-14, [Paper](https://arxiv.org/abs/2502.09992) **Dream 7B** 2025-4-2, [Paper](https://hkunlp.github.io/blog/2025/dream/) **Seed Diffusion: A Large-Scale Diffusion Language Model with High-Speed Inference** 2025-8-4, [Paper](https://www.arxiv.org/abs/2508.02193) ## Multimodal Model ### Multimodal Understanding **LLaDA-V: Large Language Diffusion Models with Visual Instruction Tuning** 2025-5-22, [Paper](https://arxiv.org/abs/2505.16933) **LaViDa: A Large Diffusion Language Model for Multimodal Understanding** 2025-5-22, [Paper](https://arxiv.org/abs/2505.16839) **Dimple: Discrete Diffusion Multimodal Large Language Model with Parallel Decoding** 2025-5-22, [Paper](https://arxiv.org/abs/2505.16990) ### Unified Multimodal Model **MMaDA: Multimodal Large Diffusion Language Models** 2025-5-21, [Paper](https://arxiv.org/abs/2505.15809) ### Speech / ASR **Whisfusion: Parallel ASR Decoding via a Diffusion Transformer** 2025-8-9, [Paper](https://arxiv.org/abs/2508.07048) ## Fast Sampling ### Enable KV Cache **dKV-Cache: The Cache for Diffusion Language Models** 2025-5-21, [Paper](https://arxiv.org/abs/2505.15781) **dLLM-Cache: Accelerating Diffusion Large Language Models with Adaptive Caching** 2025-5-22, [Paper](https://github.com/maomaocun/dLLM-cache?tab=readme-ov-file) **Accelerating Diffusion Language Model Inference via Efficient KV Caching and Guided Diffusion** 2025-5-27, [Paper](https://arxiv.org/pdf/2505.21467) **Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decoding** 2025-5-28, [Paper](https://nvlabs.github.io/Fast-dLLM/paper/fast_dllm.pdf) **Esoteric Language Models** 2025-6-2, [Paper](https://arxiv.org/abs/2506.01928) **Sparse-dLLM: Accelerating Diffusion LLMs with Dynamic Cache Eviction** 2025-8-4, [Paper](https://arxiv.org/abs/2508.02558) **DPad: Efficient Diffusion Language Models with Suffix Dropout** 2025-8-19, [Paper](https://arxiv.org/abs/2508.14148) ### Advanced Sampling Method **Dimple: Discrete Diffusion Multimodal Large Language Model with Parallel Decoding** 2025-5-22, [Paper](https://arxiv.org/abs/2505.16990) **Variational Autoencoding Discrete Diffusion with Enhanced Dimensional Correlations Modeling** 2025-5-23, [Paper](https://arxiv.org/abs/2505.17384) **Accelerating Diffusion Language Model Inference via Efficient KV Caching and Guided Diffusion** 2025-5-27, [Paper](https://arxiv.org/pdf/2505.21467) **Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decoding** 2025-5-28, [Paper](https://nvlabs.github.io/Fast-dLLM/paper/fast_dllm.pdf) **Accelerated Sampling from Masked Diffusion Models via Entropy Bounded Unmasking** 2025-5-30, [Paper](https://arxiv.org/abs/2505.24857) **Accelerating Diffusion LLMs via Adaptive Parallel Decoding** 2025-5-31, [Paper](https://arxiv.org/abs/2506.00413) **Accelerating Diffusion Large Language Models with SlowFast Sampling: The Three Golden Principles** 2025-6-12, [Paper](https://arxiv.org/abs/2506.10848) **Wide-In, Narrow-Out: Revokable Decoding for Efficient and Effective DLLMs** 2025-7-24，[Paper](https://arxiv.org/abs/2507.18578) **DPad: Efficient Diffusion Language Models with Suffix Dropout** 2025-8-19, [Paper](https://arxiv.org/abs/2508.14148) ## Reinforcement Learning **d1: Scaling Reasoning in Diffusion Large Language Models via Reinforcement Learning** 2025-4-16, [Paper](https://arxiv.org/abs/2504.12216) **Reinforcing the Diffusion Chain of Lateral Thought with Diffusion Language Models** 2025-5-15, [Paper](https://arxiv.org/abs/2505.10446) **LLaDA 1.5: Variance-Reduced Preference Optimization for Large Language Diffusion Models** 2025-5-25, [Paper](https://arxiv.org/abs/2505.19223) **DiffuCoder: Understanding and Improving Masked Diffusion Models for Code Generation** 2025-7-25, [Paper](https://arxiv.org/abs/2506.20639) **MDPO: Overcoming the Training-Inference Divide of Masked Diffusion Language Models** 2025-8-18, [Paper](https://arxiv.org/pdf/2508.13148) **Revolutionizing Reinforcement Learning Framework for Diffusion Large Language Models** 2025-9-8, [Paper](https://arxiv.org/pdf/2509.06949) **Inpainting-Guided Policy Optimization for Diffusion Large Language Models** 2025-9-12, [Paper](https://arxiv.org/pdf/2509.10396) ## Long Context **LongLLaDA: Unlocking Long Context Capabilities in Diffusion LLMs** 2025-6-17, [Paper](https://arxiv.org/abs/2506.14429) ## Variable Length **Edit Flows: Flow Matching with Edit Operations** 2025-6-10, [Paper](https://arxiv.org/abs/2506.09018) **DreamOn: Diffusion Language Models For Code Infilling Beyond Fixed-Size Canvas** 2025-7-15, [Paper](https://hkunlp.github.io/blog/2025/dreamon/) **Beyond Fixed: Variable-Length Denoising for Diffusion Large Language Models** 2025-8-4, [Paper](https://arxiv.org/abs/2508.00819) **Any-Order Flexible Length Masked Diffusion** 2025-8-31, [Paper](https://arxiv.org/pdf/2509.01025) ## Others **Time Is a Feature: Exploiting Temporal Dynamics in Diffusion Language Models** 2025-8-12, [Paper](https://arxiv.org/abs/2508.09138) **Thinking Inside the Mask: In-Place Prompting in Diffusion LLMs** 2025-8-14, [Paper](https://arxiv.org/pdf/2508.10736)