Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2510.16932 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912679045627904 |
|---|---|
| author | Xiao, Emily Zeng, Yixiao Chen, Ada Li, Chin-Jou Bertsch, Amanda Neubig, Graham |
| author_facet | Xiao, Emily Zeng, Yixiao Chen, Ada Li, Chin-Jou Bertsch, Amanda Neubig, Graham |
| contents | A popular method to adapt large language models (LLMs) to new tasks is in-context learning (ICL), which is effective but incurs high inference costs as context length grows. In this paper we propose a method to perform instruction induction, where we take training examples and reduce them to a compact but descriptive prompt that can achieve performance comparable to ICL over the full training set. Specifically, we propose PROMPT-MII, a reinforcement learning (RL) based framework to meta-learn an instruction induction model that can generate compact instructions on the fly for an arbitrary new dataset. We train on over 3,000 diverse classification datasets from the HuggingFace hub, and evaluate on 90 unseen tasks. PROMPT-MII improves downstream model quality by 4-9 F1 points (10-20% relative), matching ICL performance while requiring 3-13x fewer tokens. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_16932 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Prompt-MII: Meta-Learning Instruction Induction for LLMs Xiao, Emily Zeng, Yixiao Chen, Ada Li, Chin-Jou Bertsch, Amanda Neubig, Graham Computation and Language A popular method to adapt large language models (LLMs) to new tasks is in-context learning (ICL), which is effective but incurs high inference costs as context length grows. In this paper we propose a method to perform instruction induction, where we take training examples and reduce them to a compact but descriptive prompt that can achieve performance comparable to ICL over the full training set. Specifically, we propose PROMPT-MII, a reinforcement learning (RL) based framework to meta-learn an instruction induction model that can generate compact instructions on the fly for an arbitrary new dataset. We train on over 3,000 diverse classification datasets from the HuggingFace hub, and evaluate on 90 unseen tasks. PROMPT-MII improves downstream model quality by 4-9 F1 points (10-20% relative), matching ICL performance while requiring 3-13x fewer tokens. |
| title | Prompt-MII: Meta-Learning Instruction Induction for LLMs |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2510.16932 |