DMax uses On-Policy Uniform Training and Soft Parallel Decoding to enable aggressive parallelism in dLLMs, raising TPF on GSM8K from 2.04 to 5.47 and on MBPP from 2.71 to 5.86 while preserving accuracy.
hub Canonical reference
Dimple: Discrete diffusion multimodal large language model with parallel decoding.arXiv preprint arXiv:2505.16990
Canonical reference. 80% of citing Pith papers cite this work as background.
17
Pith papers citing it
Background
80% of classified citations
abstract
In this work, we propose Dimple, the first Discrete Diffusion Multimodal Large Language Model (DMLLM). We observe that training with a purely discrete diffusion approach leads to significant training instability, suboptimal performance, and severe length bias issues. To address these challenges, we design a novel training paradigm that combines an initial autoregressive phase with a subsequent diffusion phase. This approach yields the Dimple-7B model, trained on the same dataset and using a similar training pipeline as LLaVA-NEXT. Dimple-7B ultimately surpasses LLaVA-NEXT in performance by 3.9%, demonstrating that DMLLM can achieve performance comparable to that of autoregressive models. To improve inference efficiency, we propose a decoding strategy termed confident decoding, which dynamically adjusts the number of tokens generated at each step, significantly reducing the number of generation iterations. In autoregressive models, the number of forward iterations during generation equals the response length. With confident decoding, however, the number of iterations needed by Dimple is even only $\frac{\text{response length}}{3}$. We also re-implement the prefilling technique in autoregressive models and demonstrate that it does not significantly impact performance on most benchmark evaluations, while offering a speedup of 1.5x to 7x. Additionally, we explore Dimple's capability to precisely control its response using structure priors. These priors enable structured responses in a manner distinct from instruction-based or chain-of-thought prompting, and allow fine-grained control over response format and length, which is difficult to achieve in autoregressive models. Overall, this work validates the feasibility and advantages of DMLLM and enhances its inference efficiency and controllability. Code and models are available at https://github.com/yu-rp/Dimple.
hub tools
citation-role summary
background 4
baseline 1
citation-polarity summary
representative citing papers
Info-Gain Sampler improves MDM decoding by using bidirectional information gain to reduce cumulative uncertainty, outperforming greedy samplers on reasoning accuracy and creative writing tasks.
Joint AR–diffusion training yields one tri-mode LM that switches AR, diffusion, and self-speculation, beating open AR/diffusion models on accuracy and tokens-per-forward.
A parallel-in-time τ-leaping sampler for absorbing discrete diffusion models is introduced, with an exponential-factorial convergence proof and empirical speedups of 7-9× on synthetic tasks and 1.45-1.86× on image/text tasks while using 50% fewer NFE.
SLIM-RL matches or exceeds TraceRL performance on MATH500, GSM8K, MBPP and HumanEval for diffusion LLMs by risk-budgeted random-masking RL without trajectory slicing.
Fast-dLLM++ generalizes Fast-dLLM decoding to heterogeneous confidence profiles via Fréchet profile selection, delivering up to 37% throughput gains on GSM8K, MATH, HumanEval, and MBPP with LLaDA-8B.
dMoE aggregates token expert distributions to block level in dLLMs, cutting unique experts from 69.5 to 14.6, memory by 76-80%, and latency by 1.14-1.66x while retaining 99.11% performance.
VRCD prioritizes visually complementary positions during parallel decoding in dMLLMs by measuring attention overlap with the new Visual Redundancy Index, yielding accuracy gains over confidence-based baselines on M^3CoT and MMBench.
Fast-dDrive is a block-diffusion VLA that reports SOTA accuracy on WOD-E2E and nuScenes driving benchmarks together with 12x throughput over autoregressive baselines via section scaffolds and test-time averaging.
FullFlow adds LoRA adapters and discrete text insertion to pretrained rectified-flow text-to-image models, achieving bidirectional generation with major gains in FID, CIDEr, VRAM, and throughput over Dual Diffusion baselines.
Cola DLM proposes a hierarchical latent diffusion model that learns a text-to-latent mapping, fits a global semantic prior in continuous space with a block-causal DiT, and performs conditional decoding, establishing latent prior modeling as an alternative to token-level autoregressive language model
Progressive supervised block merging plus stage-wise intra-diffusion distillation converts Qwen3-VL into a large-block dVLM with claimed SOTA open-dVLM scores and up to 3× decoding speedup on ≤4.4M data.
Dataset-level metrics in diffusion language models mask substantial sample-level non-determinism that varies with model and system factors, which a new Factor Variance Attribution metric can decompose.
Position and step penalty plus visual reasoning guidance fix premature answering and weak visual grounding in diffusion MLLMs, delivering up to 7.5% accuracy gains and over 3x speedup.
ECHO introduces one-step block diffusion via Direct Conditional Distillation and Response-Asymmetric Diffusion to generate chest X-ray reports faster than autoregressive models while improving clinical metrics.
A token-history 'credit' score fused into current logits reduces redundant re-masking in diffusion LLM decoding, yielding up to 5.48× tokens-per-forward speedup with small accuracy gains.
Threshold-based decoding for diffusion ASR outperforms fixed schemes by accepting high-confidence tokens early and matches autoregressive accuracy with better speed.
citing papers explorer
-
DMax: Aggressive Parallel Decoding for dLLMs
DMax uses On-Policy Uniform Training and Soft Parallel Decoding to enable aggressive parallelism in dLLMs, raising TPF on GSM8K from 2.04 to 5.47 and on MBPP from 2.71 to 5.86 while preserving accuracy.
-
Improving Sampling for Masked Diffusion Models via Information Gain
Info-Gain Sampler improves MDM decoding by using bidirectional information gain to reduce cumulative uncertainty, outperforming greedy samplers on reasoning accuracy and creative writing tasks.
-
Nemotron-Labs-Diffusion: A Tri-Mode Language Model Unifying Autoregressive, Diffusion, and Self-Speculation Decoding
Joint AR–diffusion training yields one tri-mode LM that switches AR, diffusion, and self-speculation, beating open AR/diffusion models on accuracy and tokens-per-forward.
-
Accelerating Discrete Diffusion Models with Parallel-In-Time Sampling
A parallel-in-time τ-leaping sampler for absorbing discrete diffusion models is introduced, with an exponential-factorial convergence proof and empirical speedups of 7-9× on synthetic tasks and 1.45-1.86× on image/text tasks while using 50% fewer NFE.
-
SLIM-RL: Risk-Budgeted Random-Masking RL for Diffusion LLMs Without Trajectory Slicing
SLIM-RL matches or exceeds TraceRL performance on MATH500, GSM8K, MBPP and HumanEval for diffusion LLMs by risk-budgeted random-masking RL without trajectory slicing.
-
Fast-dLLM++: Fr\'{e}chet Profile Decoding for Faster Diffusion LLM Inference
Fast-dLLM++ generalizes Fast-dLLM decoding to heterogeneous confidence profiles via Fréchet profile selection, delivering up to 37% throughput gains on GSM8K, MATH, HumanEval, and MBPP with LLaDA-8B.
-
dMoE: dLLMs with Learnable Block Experts
dMoE aggregates token expert distributions to block level in dLLMs, cutting unique experts from 69.5 to 14.6, memory by 76-80%, and latency by 1.14-1.66x while retaining 99.11% performance.
-
Visual-Redundancy-Controlled Parallel Decoding for Diffusion-Based Multimodal Large Language Models
VRCD prioritizes visually complementary positions during parallel decoding in dMLLMs by measuring attention overlap with the new Visual Redundancy Index, yielding accuracy gains over confidence-based baselines on M^3CoT and MMBench.
-
Fast-dDrive: Efficient Block-Diffusion VLM for Autonomous Driving
Fast-dDrive is a block-diffusion VLA that reports SOTA accuracy on WOD-E2E and nuScenes driving benchmarks together with 12x throughput over autoregressive baselines via section scaffolds and test-time averaging.
-
FullFlow: Upgrading Text-to-Image Flow Matching Models for Bidirectional Vision--Language Generation
FullFlow adds LoRA adapters and discrete text insertion to pretrained rectified-flow text-to-image models, achieving bidirectional generation with major gains in FID, CIDEr, VRAM, and throughput over Dual Diffusion baselines.
-
Continuous Latent Diffusion Language Model
Cola DLM proposes a hierarchical latent diffusion model that learns a text-to-latent mapping, fits a global semantic prior in continuous space with a block-causal DiT, and performs conditional decoding, establishing latent prior modeling as an alternative to token-level autoregressive language model
-
BARD: Bridging AutoRegressive and Diffusion Vision-Language Models Via Highly Efficient Progressive Block Merging and Stage-Wise Distillation
Progressive supervised block merging plus stage-wise intra-diffusion distillation converts Qwen3-VL into a large-block dVLM with claimed SOTA open-dVLM scores and up to 3× decoding speedup on ≤4.4M data.
-
Dataset-Level Metrics Attenuate Non-Determinism: A Fine-Grained Non-Determinism Evaluation in Diffusion Language Models
Dataset-level metrics in diffusion language models mask substantial sample-level non-determinism that varies with model and system factors, which a new Factor Variance Attribution metric can decompose.
-
Thinking Diffusion: Penalize and Guide Visual-Grounded Reasoning in Diffusion Multimodal Language Models
Position and step penalty plus visual reasoning guidance fix premature answering and weak visual grounding in diffusion MLLMs, delivering up to 7.5% accuracy gains and over 3x speedup.
-
ECHO: Efficient Chest X-ray Report Generation with One-step Block Diffusion
ECHO introduces one-step block diffusion via Direct Conditional Distillation and Response-Asymmetric Diffusion to generate chest X-ray reports faster than autoregressive models while improving clinical metrics.
-
CreditDecoding: Accelerating Parallel Decoding in Diffusion Large Language Models with Trace Credit
A token-history 'credit' score fused into current logits reduces redundant re-masking in diffusion LLM decoding, yielding up to 5.48× tokens-per-forward speedup with small accuracy gains.
-
Decoding Strategies for Diffusion-Based ASR: A Systematic Evaluation of Confidence-Based Thresholding
Threshold-based decoding for diffusion ASR outperforms fixed schemes by accepting high-confidence tokens early and matches autoregressive accuracy with better speed.