Polestar: Drift-Aware Cache Calibration and Token Commitment for Efficient Inference of Diffusion LLMs
Quick Answer
Polestar introduces a training-free inference framework that enhances diffusion large language models (dLLMs) by addressing token representation drift.
Quick Take
It achieves up to 10.73% accuracy improvement and 3.7x higher throughput, setting new benchmarks in efficiency and decoding parallelism.
Key Points
- Polestar-Cache refreshes stale KV-cache positions for efficient reuse.
- Polestar-Commit detects drift events to identify commit-ready tokens.
- Achieves 3.67 tokens per forward pass, enhancing decoding parallelism.
- Sets new state-of-the-art on accuracy-throughput Pareto frontier.
- Demonstrates significant improvements across various dLLM families.
Paper Resources
📖 Reader Mode
~2 min readAbstract: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.
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2607.14107 [cs.CL] |
| (or arXiv:2607.14107v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2607.14107 arXiv-issued DOI via DataCite |
Submission history
From: Akshat Ramachandran [view email]
[v1]
Thu, 7 May 2026 18:05:39 UTC (3,600 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CL
See more →TriAgent: Divergence-Aware Committees for Cost-Efficient Financial Sentiment Analysis
TriAgent introduces a cost-efficient multi-agent system for financial sentiment analysis, combining VADER, FinBERT, and Qwen2.5. It achieves an F1 score of ~0.87 with significant savings of $9.3M/year at a 10M-user scale compared to GPT-4o-mini, while also detecting hallucinations with an AUC of 0.90.