Select or Project? Evaluating Lower-dimensional Vectors for LLM Training Data Explanations
This work addresses the computational bottleneck in gradient-based explanations for LLMs, offering a practical solution for researchers and practitioners, though it is incremental as it builds on existing methods with a novel evaluation.
The paper tackled the problem of high-dimensional gradients hindering instance-based explanations for large language models by comparing component selection versus projection for creating low-dimensional representations. The result showed that a greedily selected subset of components more effectively captures training data influence for retrieval tasks and is computationally efficient, outperforming full gradients and random projection.
Gradient-based methods for instance-based explanation for large language models (LLMs) are hindered by the immense dimensionality of model gradients. In practice, influence estimation is restricted to a subset of model parameters to make computation tractable, but this subset is often chosen ad hoc and rarely justified by systematic evaluation. This paper investigates if it is better to create low-dimensional representations by selecting a small, architecturally informed subset of model components or by projecting the full gradients into a lower-dimensional space. Using a novel benchmark, we show that a greedily selected subset of components captures the information about training data influence needed for a retrieval task more effectively than either the full gradient or random projection. We further find that this approach is more computationally efficient than random projection, demonstrating that targeted component selection is a practical strategy for making instance-based explanations of large models more computationally feasible.