CoPE: A Small Language Model for Steerable and Scalable Content Labeling

2025-12-19

AI Summary

To address the challenges of strategy memorization, poor generalization, and high deployment costs in content safety annotation tasks, this paper proposes a strategy-driven small language model (SLM) framework. Methodologically, we introduce *Contradictory Example Training*—a novel training paradigm that enhances the model’s understanding of nuanced safety policies; propose *Binocular Labeling*, a dual-perspective annotation method that automatically generates unambiguous, policy-aligned training data; and integrate instruction tuning with strategy-conditioned inference. In terms of contributions and results: the resulting model has only 9 billion parameters—just 1% the size of state-of-the-art large language models—yet achieves comparable or superior accuracy across seven major harm categories. It is fully deployable on a single consumer-grade GPU, significantly reducing inference latency and hardware requirements. The model and code are publicly released to foster reproducibility and community advancement.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Large Vision Models

Application Category

• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Search and Retrieval-Augmented AI: Large language models for search
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

This paper details the methodology behind CoPE, a policy-steerable small language model capable of fast and accurate content labeling. We present a novel training curricula called Contradictory Example Training that enables the model to learn policy interpretation rather than mere policy memorization. We also present a novel method for generating content policies, called Binocular Labeling, which enables rapid construction of unambiguous training datasets. When evaluated across seven different harm areas, CoPE exhibits equal or superior accuracy to frontier models at only 1% of their size. We openly release a 9 billion parameter version of the model that can be run on a single consumer-grade GPU. Models like CoPE represent a paradigm shift for classifier systems. By turning an ML task into a policy writing task, CoPE opens up new design possibilities for the governance of online platforms.

Problem

• Develops a small language model for efficient content labeling
• Introduces training to interpret policies, not just memorize them
• Enables rapid creation of clear datasets for policy training

Innovation

• Contradictory Example Training for policy interpretation
• Binocular Labeling for unambiguous dataset generation
• Small 9B parameter model for efficient GPU deployment