Research map / Large language models / Alignment and preferences
Towards Self-Robust LLMs: Intrinsic Prompt Noise Resistance via CoIPO
Xin Yang, Letian Li, Abudukelimu Wuerkaixi, Xuxin Cheng, Cao Liu, Ke Zeng, Xunliang Cai, Wenyuan Jiang
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 |
|---|---|---|---|
| train with CoIPO instead of COIN | Alpaca (Llama-7B) · NoisyPromptBench avg | accuracy (%): 58.60 -> 63.90 | Improved |
| train with CoIPO instead of COIN | Qwen2.5-7B · NoisyPromptBench avg | accuracy (%): 81.48 -> 83.45 | Improved |
| add inverse DPO to contrastive learning (ablation) | CoIPO (Llama) · NoisyPromptBench avg | accuracy (%): 58.60 -> 63.90 | Improved |
| add contrastive learning to inverse DPO (ablation) | CoIPO (Qwen) · NoisyPromptBench avg | accuracy (%): 82.86 -> 83.45 | Improved |
| use CoIPO instead of plain SFT on noisy prompts | Llama · Avg accuracy drop under perturbation | drop (%): 3.20 -> 3.88 | Hurt |
| use CoIPO instead of COIN | Qwen2.5-7B · Avg accuracy drop under perturbation | drop (%): 1.81 -> 0.54 | Improved |
| scale CoIPO to 14B and 72B | Qwen2.5 · NoisyPromptBench avg | accuracy (%): 83.45 -> 83.93 -> 85.00 | Improved |
| apply COIN at larger scale | COIN on Qwen2.5-14B · NoisyPromptBench avg | accuracy (%): 81.48 (7B) -> 71.86 (14B) | Hurt |
| train with CoIPO (not trained on these tasks) | Qwen2.5-14B · GSM8K | exact match: 75.66 -> 84.76 | Improved |
| train with CoIPO (not trained on these tasks) | Qwen2.5-7B · MBPP | Pass@1: 35.6 -> 32.8 | Hurt |
| train with CoIPO | Llama-7B · GSM8K | exact match: 7.20 -> 6.29 | Hurt |
Closest papers by meaning
- Brittlebench: Quantifying LLM robustness via prompt sensitivity (NeurIPS 2026 · similarity 0.89)
- On the Worst Prompt Performance of Large Language Models (NeurIPS 2024 · similarity 0.89)
- Align-Pro: A Principled Approach to Prompt Optimization for LLM Alignment (AAAI 2025 · similarity 0.88)
- Robust LLM Alignment via Distributionally Robust Direct Preference Optimization (NeurIPS 2025 · similarity 0.88)
- Large Language Models Must Be Taught to Know What They Don’t Know (NeurIPS 2024 · similarity 0.88)
- Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formatting (ICLR 2024 · 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.
