PlanE: Meta Planning of Data, Tuning, and Inference for Extractive-based LLMs

Short summary

PlanE is a planning framework for building extractive-based LLMs that integrates data decomposition, instruction tuning, and prompt inference. It introduces a Data-Tuning-Inference (DTI) planner that selects the optimal base-LLM and its DTI combinations for specific datasets to reduce annotation costs. Experiments validate effectiveness across different datasets and base-LLMs, with generalizability confirmed under different optimization objectives.

• •Framework for constructing extractive-based LLMs with data, tuning, and inference planning
• •DTI planner selects optimal base-LLM and configurations per dataset
• •Reduces annotation costs while improving task-specific performance

Generated with AI, which can make mistakes.