Structural Pruning for LLaMA

Towards Meta-Pruning via Optimal Transport, ICLR 2024 (Spotlight)

Code for paper "Accelerating Federated Learning for IoT in Big Data Analytics with Pruning, Quantization and Selective Updating"

A PyTorch implementation for structural pruning applied to neural networks during training

Official ICML 2026 Spotlight implementation for structural MoE compression, including attribution-guided channel scoring, coverage-maximized pruning, compact checkpoint construction, and fine-tuning support.

Multimodal-MoE-Slimming is a post-training compression framework for multimodal Mixture-of-Experts models. It analyzes how experts respond differently to visual and textual information, then uses this modality-aware behavior to reduce redundant expert capacity and build compact MoE checkpoints for more efficient deployment.

An independent reproduction of DepGraph (CVPR 2023) for ResNet-18 structural pruning. (Compression: 73.26% MACs, Accuracy: 91.69%)