Research map / Large language models / Language, safety and interpretability
Rainbow Padding: Mitigating Early Termination in Instruction-Tuned Diffusion LLMs
BumJun Kim, Dongjae Jeon, Dueun Kim, Wonje Jeung, Albert No
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 |
|---|---|---|---|
| use Rainbow Padding instead of eos padding | LLaDA-Base SFT · MATH (max_length 1024, 1 block) | accuracy: 0.9 -> 34.3 | Improved |
| use Rainbow Padding instead of eos padding | LLaDA-Base SFT · GSM8K (max_length 1024, 1 block) | accuracy: 40.2 -> 79.6 | Improved |
| use Rainbow Padding instead of eos padding | Dream-Base SFT · GSM8K (max_length 1024) | accuracy: 9.1 -> 77.3 | Improved |
| use Rainbow Padding instead of eos padding | LLaDA-Base SFT with 16 blocks · HumanEval | accuracy: 41.5 -> 39.6 | Hurt |
| use Rainbow Padding instead of eos padding | LLaDA-Base SFT · MMLU | accuracy: 64.8 -> 65.3 | No clear effect |
| use Rainbow Padding instead of eos padding | LLaDA-Base SFT · HellaSwag | accuracy: 62.5 -> 61.3 | No clear effect |
| increase number of semi-AR blocks from 1 to 16 for eos padding (ablation) | eos padding LLaDA-Base · MATH | accuracy: 0.9 -> 25.7 | Improved |
| increase number of semi-AR blocks from 1 to 16 for Rainbow Padding (ablation) | Rainbow Padding LLaDA-Base · MATH | accuracy: 34.3 -> 34.3 | No clear effect |
| add single-epoch LoRA Rainbow Padding adaptation | LLaDA instruct · MATH | accuracy: 0.1 -> 21.9 | Improved |
| add single-epoch LoRA Rainbow Padding adaptation | LLaDA instruct · GSM8K | accuracy: 3.0 -> 72.7 | Improved |
| add single-epoch LoRA Rainbow Padding adaptation | Dream instruct · MATH | accuracy: 0.0 -> 32.4 | Improved |
| use 7 pad tokens instead of 1 (ablation) | Rainbow Padding LLaDA-Base · GSM8K | accuracy: 15.9 -> 79.6 | Improved |
Closest papers by meaning
- Polestar: Drift-Aware Cache Calibration and Token Commitment for Efficient Inference of Diffusion LLMs (NeurIPS 2026 · similarity 0.90)
- The Flexibility Trap: Rethinking the Value of Arbitrary Order in Diffusion Language Models (ICML 2026 · similarity 0.89)
- Efficient Test-Time Scaling via Hierarchical Search and Self-Verification for Discrete Diffusion Language Models (ICML 2026 · similarity 0.89)
- From Next-Token to Next-Block: A Principled Adaptation Path for Diffusion LLMs (NeurIPS 2026 · similarity 0.89)
- Accelerating Diffusion LLMs via Adaptive Parallel Decoding (NeurIPS 2025 · similarity 0.89)
- SimSD: Simple Speculative Decoding in Diffusion Language Models (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.
