ssToken: Self-modulated and Semantic-aware Token Selection for LLM Fine-tuning
This addresses the need for efficient and effective fine-tuning of LLMs, offering a method that avoids reliance on additional reference models and enhances token selection, though it is incremental in the context of existing token-level approaches.
The paper tackles the problem of token-level data selection for fine-tuning large language models by proposing ssToken, which uses self-modulated loss differences and semantic-aware attention metrics to select tokens, resulting in performance improvements over full-data fine-tuning and prior methods.
Data quality plays a critical role in enhancing supervised fine-tuning (SFT) for large language models (LLMs), and token-level data selection has emerged as a promising direction for its fine-grained nature. Despite their strong empirical performance, existing token-level selection methods share two key limitations: (1) requiring training or accessing an additional reference model, and (2) relying solely on loss information for token selection, which cannot well preserve semantically important tokens that are not favored by loss-based metrics. To address these challenges, we propose ssToken, a Self-modulated and Semantic-aware Token Selection approach. ssToken leverages readily accessible history models to compute the per-token loss difference with the current model, which serves as a self-modulated signal that enables the model to adaptively select tokens along its optimization trajectory, rather than relying on excess loss from an offline-trained reference model as in prior works. We further introduce a semantic-aware, attention-based token importance estimation metric, orthogonal to loss-based selection and providing complementary semantic information for more effective filtering. Extensive experiments across different model families and scales demonstrate that both self-modulated selection and semantic-aware selection alone outperform full-data fine-tuning, while their integration--ssToken--achieves synergistic gains and further surpasses prior token-level selection methods, delivering performance improvements while maintaining training efficiency.