LLMs can Compress LLMs: Adaptive Pruning by Agents

TL;DR

This paper presents an adaptive, agent-guided pruning method for large language models that intelligently preserves critical knowledge pathways, significantly improving performance and knowledge retention at high sparsity levels without retraining.

Contribution

It introduces a novel agent-based pruning framework that combines sensitivity profiling with self-reflection, enabling effective, model-agnostic compression of LLMs while maintaining performance.

Findings

56% relative improvement in MMLU accuracy

19x better factual knowledge retention on FreebaseQA

69% lower perplexity degradation

Abstract

As Large Language Models (LLMs) continue to scale, post-training pruning has emerged as a promising approach to reduce computational costs while preserving performance. Existing methods such as SparseGPT and Wanda achieve high sparsity through layer-wise weight reconstruction or activation-aware magnitude pruning, but rely on uniform or hand-crafted heuristics to determine per-layer sparsity ratios. Moreover, recent work has shown that pruned LLMs suffer from severe factual knowledge degradation, with structured pruning methods experiencing near-total collapse in factual question-answering capabilities. We introduce agent-guided pruning, where a foundation model acts as an adaptive pruning agent to intelligently select which layers to prune at each iteration while preserving critical knowledge pathways. Our method constructs layer-wise sensitivity profiles by combining Wanda-inspired weight-activation metrics with gradient importance scores, normalized as z-scores for model-agnostic comparison. These statistics are processed by an LLM agent equipped with self-reflection capabilities, enabling it to learn from previous pruning outcomes and iteratively refine its strategy. A checkpoint rollback mechanism maintains model quality by reverting when perplexity degradation exceeds a threshold. We evaluate our approach on Qwen3 models (4B and 8B parameters) at approximately 45% sparsity, demonstrating substantial improvements over structured pruning baselines: 56% relative improvement in MMLU accuracy, 19x better factual knowledge retention on FreebaseQA, and 69% lower perplexity degradation. Notably, our framework requires no retraining, operates in a model-agnostic manner, and exhibits effective self-correction with only 2-4 rollbacks across 21-40 iterations, demonstrating that foundation models can effectively guide the compression of other foundation models.