ModelRefs / Model Pruning — AI Glossary
Model Pruning — AI Glossary
Removing low-importance weights or entire attention heads/layers from a trained model to reduce size and inference cost.
Overview
Pruning methods include unstructured (zeroing individual weights), structured (removing heads, rows, or layers), and semi-structured (2:4 sparsity for NVIDIA sparse tensor cores). Structured pruning is hardware-efficient; unstructured yields higher compression at acceptable accuracy. Often followed by quantization and knowledge distillation.
Reference details
| Topic | training |
|---|---|
| Also known as | weight pruning, network pruning |
| Last reviewed | 2026-06-24 |
Related terms
Continue your research
Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Model Pruning — AI Glossary.
Frequently asked questions
What is Model Pruning?
Removing low-importance weights or entire attention heads/layers from a trained model to reduce size and inference cost.
Is Model Pruning the same as weight pruning?
Yes — weight pruning, network pruning are common aliases for Model Pruning.
What concepts are related to Model Pruning?
Closely related concepts include quantization, model distillation, knowledge distillation.