Research map / Large language models / Language, safety and interpretability
Wide-In, Narrow-Out: Revokable Decoding for Efficient and Effective DLLMs
Feng Hong, Geng Yu, Yushi Ye, Haicheng Huang, Huangjie Zheng, Ya Zhang, Yanfeng Wang, Jiangchao Yao
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 |
|---|---|---|---|
| apply WINO draft-and-verify decoding | LLaDA · GSM8K | accuracy: 73.24 -> 75.82 | Improved |
| apply WINO | LLaDA · ARC-E | accuracy: 59.13 -> 81.19 | Improved |
| apply WINO | LLaDA · MBPP | accuracy: 36.40 -> 36.40 | No clear effect |
| apply WINO | MMaDA · Flickr30k | CIDEr: 53.67 -> 53.83 | No clear effect |
| apply WINO | MMaDA · ScienceQA | accuracy: 30.89 -> 42.24 | Improved |
| add verification module to draft-only decoding (ablation) | WINO · GSM8K | accuracy: 70.28 -> 75.82 (tau1=0.6); 72.33 -> 75.82 (tau1=0.9) | Improved |
| add verification module to draft-only decoding (ablation) | WINO · MMMU-val | accuracy: 19.89 -> 24.00 | Improved |
| naive parallel sampling: more tokens per step | LLaDA · GSM8K | accuracy: 73.24 -> 64.67 (4 tokens/step) | Hurt |
| raise drafting threshold tau1 from 0.6 to 0.9 in draft-only (ablation) | WINO · GSM8K | steps: 34.79 -> 81.39 steps | Hurt |
| apply WINO with full diffusion decoding | LLaDA · GSM8K (length 256) | accuracy: 34.34 -> 58.22 | Improved |
| increase generation length to 512 with WINO (ablation) | LLaDA · GSM8K | accuracy: 74.60 -> 79.91 | Improved |
| add shadow block for verification | WINO · GSM8K GPU memory | memory: 16.18 -> 16.57 GB | No clear effect |
Closest papers by meaning
- Free Draft-and-Verification: Toward Lossless Parallel Decoding for Diffusion Large Language Models (NeurIPS 2026 · similarity 0.90)
- Fast-dLLM++: Fr\'{e}chet Profile Decoding for Faster Diffusion LLM Inference (NeurIPS 2026 · similarity 0.90)
- PACE-dLLM: Elastic Block Decoding via Confidence Cliff Estimation for Diffusion Language Models (NeurIPS 2026 · similarity 0.88)
- SimSD: Simple Speculative Decoding in Diffusion Language Models (NeurIPS 2026 · similarity 0.87)
- Adaptive Draft-Verification for Efficient Large Language Model Decoding (AAAI 2025 · similarity 0.87)
- DMax: Aggressive Parallel Decoding for dLLMs (NeurIPS 2026 · similarity 0.87)
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.
