Research map / Image generation / Fast diffusion sampling
Rejection Mixing: Fast Semantic Propagation of Mask Tokens for Efficient DLLM Inference
Yushi Ye, Feng Hong, Huangjie Zheng, Xu Chen, Zhiyong Chen, 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 |
|---|---|---|---|
| add continuous mixing state and rejection to confidence-threshold decoding | LLaDA · GSM8K | accuracy: 73.01 -> 75.66 | Improved |
| add ReMix decoding | LLaDA · ARC-C | accuracy: 52.17 -> 66.22 | Improved |
| add ReMix decoding | MMaDA · MathVision-mini | accuracy: 12.83 -> 12.83 | No clear effect |
| add ReMix decoding | LLaDA · GSM8K (gen 512) | accuracy: 74.68 -> 77.63 | Improved |
| add ReMix decoding | LLaDA · GSM8K (gen 128) | accuracy: 59.21 -> 59.74 | Improved |
| add ReMix decoding | LLaDA · GSM8K (block 64) | accuracy: 82.56 -> 82.48 | No clear effect |
| add ReMix decoding | LLaDA · GSM8K (block 32) | accuracy: 81.73 -> 81.88 | No clear effect |
| enable mixing module (mixing ratio above 0) (ablation) | ReMix · GSM8K | accuracy | Improved |
| raise mixing ratio beyond moderate level (ablation) | ReMix · GSM8K | accuracy: beta>0.6 | Hurt |
| raise rejection threshold too high (ablation) | ReMix · GSM8K | accuracy | Hurt |
| use ReMix instead of naive parallel decoding for quality score | LLaDA · GSM8K | GPT-4o-mini quality score: 3.23 -> 4.32 | Improved |
| use ReMix instead of WINO for throughput | LLaDA · GSM8K (block 128) | tokens per second: 74.82 -> 102.47 | Improved |
Closest papers by meaning
- Accelerating Diffusion Large Language Models with SlowFast Sampling: The Three Golden Principles (ICLR 2026 · similarity 0.91)
- Fast-dLLM++: Fr\'{e}chet Profile Decoding for Faster Diffusion LLM Inference (NeurIPS 2026 · similarity 0.90)
- DAWN: Dependency-Aware Fast Inference for Diffusion LLMs (NeurIPS 2026 · similarity 0.90)
- Free Draft-and-Verification: Toward Lossless Parallel Decoding for Diffusion Large Language Models (NeurIPS 2026 · similarity 0.90)
- Mean-Field Parallel Decoding for Discrete Diffusion Language Models (NeurIPS 2026 · similarity 0.89)
- SimSD: Simple Speculative Decoding in Diffusion Language Models (NeurIPS 2026 · similarity 0.89)
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.
