Diffusion LM papers in 2026

Not an exhaustive catalog: these are core papers recorded in the public corpus. Paper counts do not measure quality or the importance of a method.

UTC: .

Core papers matching the current filters: 5.

By month and topic
MonthPapers
January0
February0
March0
April0
May0
June0
July5
August0
September0

Topics can overlap: a paper counts once in each of its topics, but only once in the total. A topic assignment does not prove quality or reproducibility.

Latest papers · Selection methodology

2026-07-30T13:04:47+00:00 · Masked / discrete diffusion, Inference acceleration, Reasoning · Source

Where and When to Commit: Candidate-Aware Decoding for Diffusion Language Models

Diffusion language models (DLMs) expose a provisional prediction at every denoising step, creating an opportunity for generation-time early exit that stops decoding before the schedule is exhausted. Existing early-exit gates decide termination from fixed-region confidence statistics or schedule-dependent rules, evidence too coarse for a decision that freezes every remaining position at once, so they fire prematurely on long chain-of-thought outputs whose answers stabilize only near the end. Adaptive sampling, the other axis of training-free acceleration, paces how quickly positions commit whil

Useful for: Not assessed · Limitation: Not assessed

2026-07-18T16:25:17+00:00 · Masked / discrete diffusion, Reasoning, Post-training · Source

Trace-Based On-Policy Distillation for Masked Diffusion Language Models

Diffusion large language models (dLLMs) are a promising alternative to autoregressive generation. However, reasoning-oriented post-training for dLLMs remains challenging. Supervised fine-tuning (SFT) for dLLMs requires dense but often off-policy masked states, while reinforcement learning (RL) relies on sparse rewards or value modeling. This paper proposes \textbf{trace-based on-policy distillation (TOPD)}, a teacher-supervised framework that transfers reasoning ability to a target dLLM without reward estimation. The key idea is to supervise a dLLM on its own denoising trajectory, focusing on

Useful for: Not assessed · Limitation: Not assessed

2026-07-16T16:57:34+00:00 · Masked / discrete diffusion, Reasoning, Post-training · Source

Mask-Aware Policy Gradients for Diffusion Language Models

Reinforcement learning has proven effective for improving reasoning in large language models, but extending it to Masked Diffusion Language Models (MDLMs) remains challenging due to the intractability of the log-likelihood estimation. Existing approaches approximate this log-likelihood by modeling only the token predictions, ignoring the order in which positions are unmasked during generation. We observe that MDLM generation involves two decisions at each step: what tokens to place at each masked position and which positions to remask. We formalize this as a two-stage action MDP, showing that

Useful for: Not assessed · Limitation: Not assessed

2026-07-08T05:48:57+00:00 · Masked / discrete diffusion, Reasoning · Source

Constrained Decoding for Diffusion Language Models via Efficient Inference over Finite Automata

Constrained decoding is essential for serving LLMs, ensuring that generated outputs follow specific structures such as JSON schema-formatted function calls. Existing systems are designed for autoregressive models and assume left-to-right generation, masking out invalid next tokens at each step. Diffusion language models, however, break this assumption: they sample multiple positions simultaneously from a fully-factorized mean-field distribution at each denoising step. In this paper, we present an exact and tractable algorithm for sampling from the constrained mean-field posterior under any con

Useful for: Not assessed · Limitation: Not assessed

2026-07-05T19:29:13+00:00 · Masked / discrete diffusion, Reasoning · Source

Don't Commit Alone: Joint Token Commitment in Diffusion Large Language Models

Diffusion large language models (dLLMs) commit multiple tokens per denoising step by decoding each selected position independently from the shared context; when those positions are dependent, the resulting factorization error is captured by conditional total correlation, which confidence-based selection cannot observe from marginals alone. We propose CoCommit, a marker-gated coordination pass that briefly defers commitment: after the usual bundle selection, a learned marker announces the commit set and the backbone's last-$n$ layers are re-applied so marked positions coordinate -- approxi

Useful for: Not assessed · Limitation: Not assessed

Public snapshot dates
Collection recorded in this snapshot
Newest publication recorded in this snapshot
Snapshot generated

These are saved observations, not the last collector attempt or a guarantee of corpus completeness.