dattri-LLM: ספריית יחודית ומציאותית לאטריבוציה של נתוני הכשרה בסקאלה של LLM

תקציר מקורי באנגליתarXiv:2609.38767v2 Announce Type: replace-cross Abstract: Training data attribution (TDA) estimates the contribution of individual training examples to model outputs. Most scalable TDA methods rely on per-example gradients, whose computation and use at LLM scale pose challenges in efficiency, compatibility, and extensibility. We introduce dattri-LLM, a TDA library that makes gradient-based attribution more practical at scale. For efficiency, dattri-LLM uses compact gradient representations and dynamically routes gradient operations based on a cost model. For compatibility, its capture mechanism collects per-example gradients from existing training loops that call backward(), without requiring changes to the loop or its configuration. This includes distributed training with DDP and FSDP and