LLMs can Compress LLMs: Adaptive Pruning by Agents
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866914255684501504 |
|---|---|
| author | Kodathala, Sai Varun Vunnam, Rakesh |
| author_facet | Kodathala, Sai Varun Vunnam, Rakesh |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_09694 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | LLMs can Compress LLMs: Adaptive Pruning by Agents Kodathala, Sai Varun Vunnam, Rakesh Computation and Language Artificial Intelligence Computer Vision and Pattern Recognition 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. |
| title | LLMs can Compress LLMs: Adaptive Pruning by Agents |
| topic | Computation and Language Artificial Intelligence Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2601.09694 |