DMDIntel ranks input tokens by how strongly their hidden-state changes project onto the dominant linear modes of a model's token trajectory, and this ranking usually matches GPT-4.1-labeled important tokens better than PCA, IG, or SHAP.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
DMDIntel: Interpreting Large Language Models via Dynamic Mode Decomposition
DMDIntel ranks input tokens by how strongly their hidden-state changes project onto the dominant linear modes of a model's token trajectory, and this ranking usually matches GPT-4.1-labeled important tokens better than PCA, IG, or SHAP.