REVIEW 31 cited by
Revolutionizing Reinforcement Learning Framework for Diffusion Large Language Models
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
Revolutionizing Reinforcement Learning Framework for Diffusion Large Language Models
read the original abstract
We propose TraceRL, a trajectory-aware reinforcement learning framework for diffusion language models (DLMs) that incorporates preferred inference trajectory into post-training, and is applicable across different architectures. Equipped with a diffusion-based value model that enhances training stability, we demonstrate improved reasoning performance on complex math and coding tasks. Besides, it can also be applied to adapt block-specific models to larger blocks, which improves sampling flexibility. Employing TraceRL, we derive a series of state-of-the-art diffusion language models, namely TraDo. Although smaller than 7B-scale AR models, TraDo-4B-Instruct still consistently outperforms them across complex math reasoning tasks. TraDo-8B-Instruct achieves relative accuracy improvements of 6.1% over Qwen2.5-7B-Instruct and 51.3% over Llama3.1-8B-Instruct on mathematical reasoning benchmarks. Through curriculum learning, we also derive the first long-CoT DLM, outperforming Qwen2.5-7B-Instruct on MATH500 with an 18.1% relative accuracy gain. To facilitate reproducible research and practical applications, we release a comprehensive open-source framework for building, training, and deploying diffusion LLMs across diverse architectures. The framework integrates accelerated KV-cache techniques and inference engines for both inference and reinforcement learning, and includes implementations of various supervised fine-tuning and RL methods for mathematics, coding, and general tasks. Code and Models: https://github.com/Gen-Verse/dLLM-RL
Forward citations
Cited by 31 Pith papers
-
Learning from the Self-future: On-policy Self-distillation for dLLMs
d-OPSD reframes on-policy self-distillation for dLLMs via suffix conditioning from self-generated answers and step-level supervision, outperforming RLVR and SFT on reasoning benchmarks with ~10% of the optimization steps.
-
Unified Energy for Invariant and Independent Decoding in Diffusion Language Models
The paper introduces Uni-E, a unified energy for DLMs that accounts for model capacity, dependency and invariance, can be computed exactly, and corrects distribution shifts from dependency and invariance.
-
Beyond Mode-Seeking RL: Trajectory-Balance Post-Training for Diffusion Language Models
TraFL applies trajectory flow balancing to post-train diffusion language models, preventing mode collapse and delivering consistent gains on reasoning tasks that hold under increased sampling.
-
Relative Score Policy Optimization for Diffusion Language Models
RSPO interprets reward advantages as targets for relative log-ratios in dLLMs, calibrating noisy estimates to stabilize RLVR training and achieve strong gains on planning tasks with competitive math reasoning performance.
-
TAD: Temporal-Aware Trajectory Self-Distillation for Fast and Accurate Diffusion LLM
TAD improves the accuracy-parallelism trade-off in diffusion LLMs via temporal-aware self-distillation that applies hard labels to soon-to-be-decoded tokens and soft supervision to future tokens.
-
MemDLM: Memory-Enhanced DLM Training
MemDLM embeds a simulated denoising trajectory into DLM training via bi-level optimization, creating a parametric memory that improves convergence and long-context performance even when the memory is dropped at test time.
-
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.
-
DICE: Diffusion Large Language Models Excel at Generating CUDA Kernels
DICE, a family of diffusion LLMs trained with a bi-phase infilling-to-generation RL curriculum, achieves state-of-the-art CUDA kernel generation at data and token budgets far below prior specialized autoregressive models.
-
The Flexibility Trap: Rethinking the Value of Arbitrary Order in Diffusion Language Models
Arbitrary-order generation lowers the reasoning Pass@k of diffusion LLMs; constraining RL to autoregressive order (JustGRPO) improves accuracy and preserves parallel decoding.
-
From Interface to Inference: Eliciting Any-Order Inference from Any-Order Models
Masked diffusion language models fail to exploit their any-order interface because of positional uncertainty; insertion-based (FlexMDM) and latent-segment (LatentMDM) variants recover distinct any-order inference beha...
-
AdaFlash: Adaptive Speculative Decoding via On-Policy Distilled Diffusion Drafters
AdaFlash improves diffusion-drafter speculative decoding with on-policy reverse-KL distillation and an online adaptive length head, sustaining acceleration under high concurrency.
-
Trace-Based On-Policy Distillation for Masked Diffusion Language Models
TOPD distills a stronger diffusion LLM into a weaker one on the student's own denoising trajectories and matches a reward-trained 4B model on MATH500 with 4x fewer rollouts.
-
Spec-AUF: Accept-Until-Fail Training under Train-Inference Misalignment for Masked Block Drafters
Accept-Until-Fail training improves average accepted block length in speculative decoding from 2.40 to 2.61 by limiting cross-entropy support to the drafter's first predicted failure point.
-
Diffusion-GR2: Diffusion Generative Reasoning Re-ranker
CFT plus on-policy distillation and RL converts an AR reasoning re-ranker into a block-diffusion model that matches AR accuracy at 2.4–3.5× higher throughput on Amazon Beauty.
-
Diffusion-GR2: Diffusion Generative Reasoning Re-ranker
Diffusion-GR2 converts an AR reasoning re-ranker to block-diffusion via CFT, OPD, and RL stages, recovering near-parity accuracy on Amazon Beauty with 2.4-3.5x decode speedup.
-
Diffusion-GR2: Diffusion Generative Reasoning Re-ranker
CFT + on-policy distillation + RL converts an AR reasoning re-ranker into a block-diffusion model that recovers near-AR accuracy at 2.4–3.5× decode throughput on Amazon Beauty.
-
HERALD: High-Throughput Block Diffusion LLM Serving via CPU-GPU Cooperative KV Cache Retrieval
HERALD enables near-lossless accuracy at 5-10% KV budget for block dLLMs by amortizing top-k selection across denoising steps and overlapping CPU-GPU retrieval, yielding up to 2.47x higher throughput than GPU-only inference.
-
VoidPadding: Let [VOID] Handle Padding in Masked Diffusion Language Models so that [EOS] Can Focus on Semantic Termination
VoidPadding decouples padding from termination in MDLMs via a new [VOID] token, delivering +17.84 average benchmark points and 55.7% fewer decoding steps on Dream-7B-Instruct.
-
Beyond Fully Random Masking: Attention-Guided Denoising and Optimization for Diffusion Language Models
AGDO improves dLLM reasoning performance by determining denoising order and emphasizing tokens based on attention-derived dependencies rather than random masking.
-
Back on Track: Aligning Rewards and States for Reasoning in Diffusion Large Language Models
PAPO improves reasoning performance in diffusion LLMs by converting sparse terminal rewards into dense step-wise credit and replaying real high-uncertainty trajectories, reporting gains up to 42.2% on Countdown.
-
Data-Efficient Autoregressive-to-Diffusion Language Models via On-Policy Distillation
On-policy distillation from a frozen autoregressive teacher to a bidirectional student eliminates train-inference mismatch and enables data-efficient ARLM-to-DLM conversion.
-
GDSD: Reinforcement Learning as Guided Denoiser Self-Distillation for Diffusion Language Models
GDSD reduces RL for dLLMs to likelihood-free self-distillation via a normalization-free logit-matching objective, outperforming ELBO methods with more stable training on LLaDA-8B and Dream-7B.
-
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.
-
Sketch Then Paint: Hierarchical Reinforcement Learning for Diffusion Multi-Modal Large Language Models
Proposes HT-GRPO with sketch-then-paint staged updates, prompt-conditioned importance ratios, and hierarchical credit assignment for dMLLMs, reporting gains on GenEval and DPG plus quality metrics.
-
Layer-Parallel Inference Reduces Encrypted Nonlinear Depth in Transformers
SNLP reduces encrypted Transformer nonlinear depth from L sequential stages to (L−N)+K, cutting symbolic bootstraps ~2.65× with lower error amplification than sequential inference.
-
Layer-Parallel Inference Reduces Encrypted Nonlinear Depth in Transformers
SNLP reduces symbolic FHE bootstraps from 53 to 20 on a 0.5B model with +1.2% PPL degradation and lower polynomial-error amplification than sequential inference.
-
Dynamic Infilling Anchors for Format-Constrained Generation in Diffusion Large Language Models
DIA is a training-free method that dynamically adjusts anchor positions in diffusion LLMs to improve format compliance and accuracy on reasoning benchmarks like GSM8K and MATH.
-
Fast-dDrive: Efficient Block-Diffusion VLM for Autonomous Driving
Fast-dDrive is a block-diffusion VLA that reports SOTA ADE on WOD-E2E, 0.32 m L2 on nuScenes, and 12x throughput over AR baselines via section scaffolds and test-time rollout averaging.
-
FlashBlock: Attention Caching for Efficient Long-Context Block Diffusion
Block-external attention is stable across diffusion steps, so caching and reusing it while recomputing only block-internal attention speeds up long-context generation with limited quality loss.
-
T$^\star$: Progressive Block Scaling for Masked Diffusion Language Models Through Trajectory Aware Reinforcement Learning
A progressive RL-based curriculum, T⋆, doubles block sizes in masked diffusion LMs while keeping or improving math reasoning accuracy.
-
Beyond Execution: Static-Analysis Rewards and Hint-Conditioned Diffusion RL for Code Generation
Static checking rewards and moderate AST-based hints improve diffusion RL performance for code generation, with effectiveness varying by task difficulty across HumanEval, MBPP, and LiveCodeBench.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.