Token-wise Influential Training Data Retrieval for Large Language Models
This addresses the need for interpretability in LLMs by enabling efficient training data attribution, though it is an incremental improvement on existing influence estimation methods.
The paper tackles the problem of identifying which training data influenced specific LLM generations by proposing RapidIn, a scalable framework that compresses gradient vectors by over 200,000x and achieves a 6,326x speedup in retrieval time.
Given a Large Language Model (LLM) generation, how can we identify which training data led to this generation? In this paper, we proposed RapidIn, a scalable framework adapting to LLMs for estimating the influence of each training data. The proposed framework consists of two stages: caching and retrieval. First, we compress the gradient vectors by over 200,000x, allowing them to be cached on disk or in GPU/CPU memory. Then, given a generation, RapidIn efficiently traverses the cached gradients to estimate the influence within minutes, achieving over a 6,326x speedup. Moreover, RapidIn supports multi-GPU parallelization to substantially accelerate caching and retrieval. Our empirical result confirms the efficiency and effectiveness of RapidIn.