AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

A Unified Approach to Interpreting Knowledge Distillation for Large Language Models via Interactions

Qingzhuo Wang, Ruiyang Qin, Zhenxin Qin, Wen Shen, Zhihua Wei

ICML 2026 · Large language models › Language, safety and interpretability · approach: Faithfulness of LLM explanations · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
distill the student with KD instead of leaving it undistilled (interaction sparsity)GPT-2, OPT, LLaMA students · Dolly test interactionsGini coefficient / entropyImproved
standard SFT instead of KD (ablation)student models · Dolly test interactionsinteraction sparsityNo clear effect
apply KD to the studentGPT-2-0.1B student · Dolly test, overlap@0.1 with teacheroverlap rate: 19.81% -> 24.84% (KD)Improved
apply DISTILLM to LLaMA-7B studentLLaMA-7B · Dolly test, overlap@0.1overlap rate: 21.35% -> 31.60%Improved
higher sparsity of complex interactions across KD methodsKD methods · Dolly, GPT-2 and OPT studentsROUGE-L correlation: Pearson 0.81 (+-0.11) complex vs 0.34 (+-0.12) simpleImproved
add Complex Interaction Penalty (CIP) loss to GKDGKD, GPT-2-0.3B · Dolly testROUGE-L: 26.35 -> 26.52Improved
add CIP loss to DISTILLMDISTILLM, GPT-2-0.3B · Dolly testROUGE-L: 27.47 -> 27.67Improved
add CIP loss to GKDGKD, OPT-1.3B · Dolly testROUGE-L: 26.00 -> 26.38Improved
add CIP loss to DISTILLMDISTILLM, OPT-1.3B · Dolly testROUGE-L: 27.69 -> 27.77Improved
add generic L1 penalty instead of CIP (ablation)GKD and DISTILLM · Dolly test, GPT-2-0.3B and OPT-1.3BROUGE-L: -1.17 to -0.56Hurt
add Tf-KD regularization instead of CIP (ablation)GKD and DISTILLM · Dolly test, GPT-2-0.3B and OPT-1.3BROUGE-L: -0.47 to -0.04Hurt
increase CIP weight lambda (ablation)CIP on GPT-2-0.3B · Dolly testROUGE-LMixed

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy