Research map / Large language models / Language, safety and interpretability
DSB: Dynamic Sliding Block Scheduling for Diffusion LLMs
Lizhuo Luo, Shenggui Li, Yonggang Wen, Tianwei Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add DSB Cache prefix-window refresh instead of Dual Cache (ablation) | DSB (const.) · GSM8K, LLaDA-8B-Instruct | accuracy: 76.42 -> 80.14 | Improved |
| add DSB Cache prefix-window refresh instead of Dual Cache (ablation) | DSB (const.) · GSM8K, LLaDA-8B-Instruct | TPS: 78.93 -> 98.10 | Improved |
| add DSB Cache prefix-window refresh instead of Dual Cache (ablation) | DSB (const.) · HumanEval, LLaDA-8B-Instruct | accuracy: 28.05 -> 37.80 | Improved |
| add DSB Cache prefix-window refresh instead of Dual Cache (ablation) | DSB (greedy) · HumanEval, LLaDA-8B-Instruct | accuracy: 28.05 -> 39.63 | Improved |
| add DSB Cache prefix-window refresh instead of Dual Cache (ablation) | DSB (greedy) · GSM8K, LLaDA-8B-Instruct | accuracy: 76.35 -> 80.29 | Improved |
| use DSB (const.) instead of AdaBlock | confidence decoding · HumanEval, LLaDA-8B-Instruct | accuracy: 39.63 -> 42.07 | Improved |
| use DSB (greedy) instead of naive block | confidence decoding · GSM8K, LLaDA-8B-Instruct | accuracy: 77.26 -> 78.54 | Improved |
| use DSB (greedy) instead of naive block | confidence decoding · GSM8K, LLaDA-8B-Instruct | TPS: 48.70 -> 51.03 | Improved |
| use DSB with DSB Cache instead of naive block with Dual Cache | confidence decoding with cache · GSM8K, LLaDA-8B-Instruct | accuracy: 77.40 -> 80.29 | Improved |
| use DSB with DSB Cache instead of naive block with Dual Cache | confidence decoding with cache · GSM8K, LLaDA-8B-Instruct | TPS: 92.26 -> 99.61 | Improved |
| use larger initial block length of 64 for DSB (ablation) | DSB · GSM8K, LLaDA-8B-Instruct | accuracy | Hurt |
| increase max block length (ablation) | DSB · LLaDA models | accuracy | Mixed |
Closest papers by meaning
- Dynamically Structured Diffusion Language Model Decoding via Bayesian Inference (NeurIPS 2026 · similarity 0.89)
- Accelerating Diffusion LLMs via Adaptive Parallel Decoding (NeurIPS 2025 · similarity 0.89)
- PACE-dLLM: Elastic Block Decoding via Confidence Cliff Estimation for Diffusion Language Models (NeurIPS 2026 · similarity 0.89)
- DAWN: Dependency-Aware Fast Inference for Diffusion LLMs (NeurIPS 2026 · similarity 0.88)
- From Next-Token to Next-Block: A Principled Adaptation Path for Diffusion LLMs (NeurIPS 2026 · similarity 0.88)
- Polestar: Drift-Aware Cache Calibration and Token Commitment for Efficient Inference of Diffusion LLMs (NeurIPS 2026 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
