IG-Pruning: Input-Guided Block Pruning for Large Language Models

Kangyu Qiao, Shaolei Zhang, Yang Feng*

Source code for paper "IG-Pruning: Input-Guided Block Pruning for Large Language Models".

If you find this project useful, feel free to ⭐️ it and give it a citation!

Quick Start

conda env create igpruning python=3.10

git clone https://github.com/ictnlp/IG-Pruning.git
cd IG-Pruning
pip install -r requirements.txt

Mask Candidate Discovery

A semantic clustering-based mask discovery stage that identifies diverse, high-quality mask candidates while capturing global information through rapidly converging trainable masks.

bash scripts/main_result/sample_cluster.sh

Dynamic Routing at Inference

A lightweight inference-time routing mechanism that requires no additional training of the base model parameters, enabling efficient dynamic adaptation to varying inputs.

bash scripts/main_result/sample_cluster.sh

Evaluation

To evaluate the performace, we use lm-evaluation-harness.

bash scripts/main_result/eval_mask.sh

Licence

This project is licensed under the Apache License, Version 2.0. See LICENSE for the full license text.

Citation

If this repository is useful for you, please cite as:

@article{qiao2025ig,
  title={IG-Pruning: Input-Guided Block Pruning for Large Language Models},
  author={Qiao, Kangyu and Zhang, Shaolei and Feng, Yang},
  journal={arXiv preprint arXiv:2511.02213},
  year={2025}
}

If you have any questions, feel free to contact qiaokangyu24s@ict.ac.cn.