ALinFiK: Learning to Approximate Linearized Future Influence Kernel for Scalable Third-Party LLM Data Valuation
This work addresses the need for scalable data valuation to optimize LLM performance within budget constraints, benefiting data providers and model developers, but it is incremental as it builds on existing data valuation methods.
The paper tackles the problem of valuing individual data samples for large language model (LLM) training by introducing a linearized future influence kernel (LinFiK) and a learning strategy (ALinFiK) to approximate it, resulting in a scalable approach that outperforms existing baselines in effectiveness and efficiency.
Large Language Models (LLMs) heavily rely on high-quality training data, making data valuation crucial for optimizing model performance, especially when working within a limited budget. In this work, we aim to offer a third-party data valuation approach that benefits both data providers and model developers. We introduce a linearized future influence kernel (LinFiK), which assesses the value of individual data samples in improving LLM performance during training. We further propose ALinFiK, a learning strategy to approximate LinFiK, enabling scalable data valuation. Our comprehensive evaluations demonstrate that this approach surpasses existing baselines in effectiveness and efficiency, demonstrating significant scalability advantages as LLM parameters increase.