Research map / Image generation / Fast diffusion sampling
DiffuCoder: Understanding and Improving Masked Diffusion Models for Code Generation
Shansan Gong, Ruixiang ZHANG, Huangjie Zheng, Jiatao Gu, Navdeep Jaitly, Lingpeng Kong, Yizhe 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 |
|---|---|---|---|
| apply coupled-GRPO RL after instruction tuning | DiffuCoder-Instruct · EvalPlus | pass@1 (%): 63.6 -> 67.9 | Improved |
| apply coupled-GRPO RL after instruction tuning | DiffuCoder-Instruct · BigCodeBench-C Hard | pass@1 (%): 12.2 -> 10.8 | Hurt |
| use complementary (coupled) mask sampling instead of decoupled random masks (ablation) | coupled-GRPO · HumanEval+ | pass@1 (%): 62.8 -> 68.3 | Improved |
| use coupled sampling instead of full-mask completion probability estimate (ablation) | coupled-GRPO · HumanEval | pass@1 (%): 66.5 -> 73.2 | Improved |
| use leave-one-out advantage baseline (ablation) | coupled-GRPO · HumanEval+ | pass@1 (%): 68.3 -> 62.2 | Mixed |
| raise sampling temperature from 0.2 to 1.0-1.2 (ablation) | DiffuCoder · HumanEval | pass@k | Improved |
| apply coupled-GRPO | DiffuCoder-Instruct · HumanEval (half decoding steps) | performance drop | Improved |
| extend adaptation pre-training from 65B to 700B tokens (ablation) | DiffuCoder · downstream validation | performance | Hurt |
| apply standard GRPO to the AR model | Qwen2.5-Coder+SFT · HumanEval | pass@1 (%): 82.9 -> 80.5 | Mixed |
Closest papers by meaning
- SimSD: Simple Speculative Decoding in Diffusion Language Models (NeurIPS 2026 · similarity 0.90)
- Efficient Test-Time Scaling via Hierarchical Search and Self-Verification for Discrete Diffusion Language Models (ICML 2026 · similarity 0.90)
- d2: Improving Reasoning in Diffusion Language Models via Trajectory Likelihood Estimation (ICML 2026 · similarity 0.90)
- PACE-dLLM: Elastic Block Decoding via Confidence Cliff Estimation for Diffusion Language Models (NeurIPS 2026 · similarity 0.89)
- Confidence-Based Decoding is Provably Efficient for Diffusion Language Models (NeurIPS 2026 · similarity 0.89)
- Reinforcement Learning for Diffusion LLMs with Entropy-Guided Step Selection and Stepwise Advantages (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.
