RoPE-LIME: RoPE-Space Locality + Sparse-K Sampling for Efficient LLM Attribution

TL;DR

RoPE-LIME is a novel explanation method for closed-source LLMs that uses a local kernel in RoPE space and sparse sampling to generate more accurate token attributions efficiently.

Contribution

It introduces RoPE-LIME, combining RoPE-based similarity and sparse sampling to improve LLM attribution efficiency and accuracy without requiring model gradients.

Findings

Outperforms leave-one-out sampling in attribution quality

Reduces API calls compared to gSMILE

Effective on HotpotQA and MMLU datasets

Abstract

Explaining closed-source Large Language Model (LLM) outputs is challenging because API access prevents gradient-based attribution, while perturbation methods are costly and noisy when they depend on regenerated text. We introduce \textbf{Rotary Positional Embedding Linear Local Interpretable Model-agnostic Explanations (RoPE-LIME)}, an open-source extension of gSMILE that decouples reasoning from explanation: given a fixed output from a closed model, a smaller open-source surrogate computes token-level attributions from probability-based objectives (negative log-likelihood and divergence targets) under input perturbations. RoPE-LIME incorporates (i) a locality kernel based on Relaxed Word Mover's Distance computed in \textbf{RoPE embedding space} for stable similarity under masking, and (ii) \textbf{Sparse-$K$} sampling, an efficient perturbation strategy that improves interaction coverage under limited budgets. Experiments on HotpotQA (sentence features) and a hand-labeled MMLU subset (word features) show that RoPE-LIME produces more informative attributions than leave-one-out sampling and improves over gSMILE while substantially reducing closed-model API calls.