AI Paper Atlas

Research map / Large language models / Reasoning and chain of thought

SKIM: Pruning Large Language Model Agents via Selective Knowledge Informed Masking

Moonseok Choi, Giung Nam, Jongwon Jeong, Minki Kang, Juho Lee

NeurIPS 2026 · Large language models › Reasoning and chain of thought · approach: Test-time compute scaling · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes SKIM, a pruning method for LLM agents under grammar-constrained structured generation that accounts for the mismatch between constrained output distributions and standard unconstrained pruning criteria.

Problem

Compression methods for LLMs target unconstrained generation, but LLM agents are often deployed under grammar-constrained structured generation, where the output distribution is restricted to valid tokens and the logits that matter lie in a small set of positions.

Approach

Identifies a mismatch under structured generation: grammar constraints reduce the output distribution to valid tokens, so logits matter most at a small set of constrained positions; SKIM uses selective, knowledge-informed masking to prune the model accordingly, rather than using unconstrained pruning criteria.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add on-policy distillation phase (ablation)SKIM · BFCL-v3 25% sparsity, Qwen3.5-2Bstructured accuracy: 78.1 -> 80.8Improved
add category-aware feature distillation (ablation)SKIM · BFCL-v3 25% sparsity, Qwen3.5-2Bstructured accuracy: 79.1 -> 80.8Improved
add vanilla output-matching KD (beta 0.5) (ablation)SKIM · BFCL-v3 25% sparsity, Qwen3.5-2Bstructured accuracy: 80.8 -> 77.9Hurt
add vanilla KD (beta 0.5) (ablation)SKIM · BFCL-v3 25% sparsity, Qwen3.5-2Bfree-generation accuracy: 0.0 -> 46.4Improved
down-weight format positions (alpha_f 0.2 instead of 1.0 or 0) (ablation)SKIM · API-Bank L1 25% sparsity, Qwen3.5-2Bstructured accuracyImproved
use tool-span rule labels instead of GPT-4 context labels (ablation)SKIM · BFCL-v3, Qwen3.5-9Bstructured accuracy: 85.5 -> 84.0 (25%), 77.6 -> 73.0 (50%)Hurt
use tool-span rule labels instead of GPT-4 context labels (ablation)SKIM · BFCL-v3, Qwen3.5-2Bstructured accuracy: 80.8 -> 81.7 (25%), 64.5 -> 64.3 (50%)No clear effect
label all prompt tokens as context (ablation)SKIM · BFCL-v3, Qwen3.5-9Bstructured accuracy: 85.5 -> 78.0 (25%)Hurt
category-aware pruning and distillation vs uniform saliency plus vanilla KDSKIM · Spider, Qwen3.5-2B 10% sparsitycomponent F1: 53.1 -> 56.4Improved
category-aware pruning and distillation vs uniform saliency plus vanilla KDSKIM · GSM8K JSON, Qwen3.5-2B 10% sparsityexact match: 21.0 -> 24.0Improved
SKIM instead of strongest baseline pruning at 50% sparsitySKIM · BFCL-v3, Hammer2.1-3Bstructured accuracy: 73.0 -> 82.8Improved
SKIM instead of baselines at 10% sparsitySKIM · BFCL-v3, Qwen3.5-2Bstructured accuracy: FLAP 83.8 vs SKIM 82.9No clear effect

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