Research
SPICE: Submodular Penalized Information-Conflict Selection for Efficient Large Language Model Training
arXiv:2601.23155v2 Announce Type: replace-cross Abstract: Information-based data selection for instruction tuning is compelling: maximizing the log-determinant of the Fisher information yields a monot
arXiv:2601.23155v2 Announce Type: replace-cross Abstract: Information-based data selection for instruction tuning is compelling: maximizing the log-determinant of the Fisher information yields a monotone submodular objective, enabling greedy algorithms to achieve a (1-1/e) approximation under a cardinality budget. In practice, however, we identify alleviating gradient conflicts, misalignment between per-sample gradients, is a key factor that slows down the decay of marginal log-determinant information gains, thereby preventing significant loss of information. We formalize this via an arepsilon-decomposition that quantifies the deviation from ideal submodularity as a function of conflict statistics, yielding data-dependent approximation factors that tighten as conflicts diminish. Guided by this analysis, we propose SPICE, a conflict-aware selector that maximizes information while penalizing misalignment, and that supports early stopping and proxy models for efficiency. Empirically, SPICE selects subsets with higher log-determinant information than original criteria, and these informational gains translate into performance improvements: across 8 benchmarks with LLaMA2-7B and Qwen2-7B, SPICE uses only 10% of the data, yet matches or exceeds 6 methods including full-data tuning. This achieves performance improvements with substantially lower training cost.
Related
- On the Step Length Confounding in LLM Reasoning Data Selection
- WRAP++: Web discoveRy Amplified Pretraining
- Less is More: Data-Efficient Adaptation for Controllable Text-to-Video Generation
- Stabilizing Unsupervised Self-Evolution of MLLMs via Continuous Softened Retracing reSampling
Source: arXiv cs.AI | 2026-04-10