mlsanityData to 14 Oct 2026

Polestar: Drift-Aware Cache Calibration and Token Commitment for Efficient Inference of Diffusion LLMs

arXiv 2607.14107 ↗First posted 7 May 2026cs.CL cs.AIcited by 1

On the map: Efficient LLM inference and serving → Diffusion LLM and MoE inference

Abstract (arXiv)

The inference efficiency of diffusion large language models (dLLMs) is constrained by two challenges: bidirectional attention precludes efficient KV-cache reuse, while increasing decoding parallelism with static confidence thresholds can compromise generation quality. We observe that both challenges arise from a shared phenomenon: as tokens are decoded, their contextual integration through bidirectional attention causes token representations to drift (evolve) across decoding steps. This insight motivates Polestar, a training-free inference framework that uses token representation drift as a unified signal to jointly address both challenges. Polestar comprises two components: Polestar-Cache, which identifies stale KV-cache positions via drift and performs sparse KV-cache refreshes to enable efficient reuse, and Polestar-Commit, which detects sharp drift events to reliably identify commit-ready tokens. Across mathematics and coding benchmarks on several dLLM families, Polestar sets a new state of the art on the accuracy-throughput Pareto frontier, achieving up to 10.73% accuracy improvement, up to 3.7x higher throughput, and high decoding parallelism of 3.67 tokens per forward pass over existing baselines.

Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Placed in Diffusion LLM and MoE inference, one of 12 directions in Efficient LLM inference and serving. Builds on 37 of its 43 references that are in the map. First posted on arXiv 7 May 2026.

Builds on

37 of 43 references.

Closest ancestors · 11

Same direction, newest first

  1. 2026DAWN: Dependency-Aware Fast Inference for Diffusion LLMs6 Feb
  2. 2025Attention Is All You Need for KV Cache in Diffusion LLMs16 Oct
  3. 2025ParallelBench: Understanding the Trade-offs of Parallel Decoding in Diffusion LLMs6 Oct
  4. 2025AdaBlock-dLLM: Semantic-Aware Diffusion LLM Inference via Adaptive Block Size30 Sep
  5. 2025Learning to Parallel: Accelerating Diffusion Large Language Models via Learnable Parallel Decoding29 Sep
  6. 2025d2Cache: Accelerating Diffusion-Based LLMs via Dual Adaptive Caching27 Sep
  7. 2025Sparse-dLLM: Accelerating Diffusion LLMs with Dynamic Cache Eviction4 Aug
  8. 2025Accelerating Diffusion LLMs via Adaptive Parallel Decoding31 May
  9. 2025Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decoding28 May
  10. 2025Dimple: Discrete Diffusion Multimodal Large Language Model with Parallel Decoding22 May
  11. 2025dKV-Cache: The Cache for Diffusion Language Models21 May

Other references · 26

  1. 2024The Llama 3 Herd of Modelscited by 14,715
  2. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  3. 2021Evaluating Large Language Models Trained on Codecited by 8,320
  4. 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,371
  5. 2021Program Synthesis with Large Language Modelscited by 3,637
  6. 2021Structured Denoising Diffusion Models in Discrete State-Spacescited by 1,719
  7. 2023MathVista: Evaluating Mathematical Reasoning of Foundation Models in Visual Contextscited by 1,614
  8. 2025Large Language Diffusion Modelscited by 912
  9. 2024MathVerse: Does Your Multi-modal LLM Truly See the Diagrams in Visual Math Problems?cited by 586
  10. 2025Dream 7B: Diffusion Large Language Modelscited by 544
  11. 2025Block Diffusion: Interpolating Between Autoregressive and Diffusion Language Modelscited by 409
  12. 2025Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attentioncited by 407
14 more
  1. 2025LLaDA 1.5: Variance-Reduced Preference Optimization for Large Language Diffusion Modelscited by 285
  2. 2025dLLM-Cache: Accelerating Diffusion Large Language Models with Adaptive Cachingcited by 169
  3. 2025LLaDA-V: Large Language Diffusion Models with Visual Instruction Tuningcited by 148
  4. 2025FlashDLM: Accelerating Diffusion Language Model Inference via Efficient KV Caching and Guided Diffusioncited by 77
  5. 2025Diffusion Language Models Know the Answer Before Decodingcited by 70
  6. 2025KLASS: KL-Guided Fast Inference in Masked Diffusion Modelscited by 47
  7. 2025Accelerating Diffusion Large Language Models with SlowFast Sampling: The Three Golden Principlescited by 45
  8. 2024MicroScopiQ: Accelerating Foundational Models through Outlier-Aware Microscaling Quantizationcited by 28
  9. 2025ThinKV: Thought-Adaptive KV Cache Compression for Efficient Reasoning Modelscited by 21
  10. 2025Multipole Attention for Efficient Long Context Reasoningcited by 14
  11. 2025Beyond Confidence: Adaptive and Coherent Decoding for Diffusion Language Modelscited by 9
  12. 2025Improving the Throughput of Diffusion-based Large Language Models via a Training-Free Confidence-Aware Calibrationcited by 7
  13. 2025SkipKV: Selective Skipping of KV Generation and Storage for Efficient Inference with Large Reasoning Modelscited by 5
  14. 2026EntropyCache: Decoded Token Entropy Guided KV Caching for Diffusion Language Modelscited by 3

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026Affix Cache for Diffusion Large Language Modelscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Flash-dLLM: IO-Aware KV Caching and Parallel Decoding for Fast, Memory-Efficient Diffusion LLMs
  2. 2026CAI-DLLM: Convergence Aware Inference for Diffusion Language Models
  3. 2026GroupKV: Hierarchical KV Cache Management for Long-Context Diffusion LLM Inference
  4. 2026Roll Out and Roll Back: Diffusion LLMs are Their Own Efficiency Teachers
  5. 2025WeDLM: Reconciling Diffusion Language Models with Standard Causal Attention for Fast Inference
  6. 2025Accelerating Diffusion LLM Inference via Local Determinism Propagation
  7. 2026Mosaic: Unlocking Long-Context Inference for Diffusion LLMs via Global Memory Planning and Dynamic Peak Taming
  8. 2026Beyond Scattered Acceptance: Fast and Coherent Inference for DLMs via Longest Stable Prefixes