Research map / Image generation / Fast diffusion sampling
Exploring More to Solve More: Boosting Diversity in Text Diffusion Models via Entropy-Based Guidance
Jingwei Zhang, Haoyu Lei, Zijin Feng, Jiacheng Sun, Farzan Farnia
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 SAKE diversity guidance | LLaDA-8B · HumanEval Pass@32 | Pass@32: 41.1 -> 55.8 | Improved |
| add SAKE diversity guidance | LLaDA-8B · MBPP Pass@32 | Pass@32: 48.2 -> 56.1 | Improved |
| add SAKE diversity guidance | LLaDA-8B · GSM8K strict self-consistency | accuracy: 71.5 -> 75.1 | Improved |
| add SAKE diversity guidance | LLaDA-8B · GSM8K flexible self-consistency | accuracy: 68.8 -> 73.5 | Improved |
| add SAKE diversity guidance | LLaDA-8B · HumanEval Pass@1 | Pass@1: 32.9 -> 32.0 | No clear effect |
| add SAKE diversity guidance | LLaDA-8B · MBPP Pass@1 | Pass@1: 40.2 -> 39.0 | Hurt |
| use SAKE instead of D-CFG | LLaDA-8B · HumanEval Pass@32 | Pass@32: 41.2 -> 55.8 | Improved |
| use D-CFG instead of base sampling | LLaDA-8B · MBPP Pass@32 | Pass@32: 48.2 -> 45.9 | Hurt |
| add SAKE guidance | LLaDA-8B · generation speed average | tokens/sec: 47.41 -> 44.09 | Hurt |
| add D-CFG guidance | LLaDA-8B · generation speed average | tokens/sec: 47.41 -> 31.83 | Hurt |
| raise top-p from 0.5 to 0.7 | LLaDA-8B base · HumanEval Pass@32 | Pass@32: 35.4 -> 40.9 | Improved |
| increase semantic bandwidth sigma from 20 to 120 (ablation) | SAKE on MDLM · Gen PPL | Gen PPL: 10.13 -> 18.33 | Hurt |
Closest papers by meaning
- Denoising Time Matters: Diverse Generation in Diffusion Language Models (NeurIPS 2026 · similarity 0.88)
- SemanticDLM+: Improving Diffusion LLMs through Bias-variance Trade-off in Transition Kernel Design (NeurIPS 2026 · similarity 0.88)
- EntropyCache: Decoded Token Entropy Guided KV Caching for Diffusion Language Models (NeurIPS 2026 · similarity 0.88)
- Discrete Diffusion Modeling by Estimating the Ratios of the Data Distribution (ICML 2024 · similarity 0.88)
- Free Lunch for Pass@$k$? Low Cost Diverse Sampling for Diffusion Language Models (NeurIPS 2026 · similarity 0.87)
- Breaking AR’s Sampling Bottleneck: Provable Acceleration via Diffusion Language Models (NeurIPS 2025 · 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.
