AI Glossary Term

What does “Pruning” mean?

Definition

Removing less important model weights or neurons to reduce size and compute.

Browse the full AI Glossary Explore all free AI guides

How Pruning works

Example

After training, an engineer finds that many weights are close to zero and contribute little. Removing them shrinks the model, and a short round of further training recovers most of the lost accuracy.

Why it matters

Pruning cuts model size and sometimes cost. Gains depend on hardware that can exploit the sparsity.

Common questions

Does pruning lower accuracy?

A little, if done too aggressively. Moderate pruning often barely changes accuracy.

Pruning vs quantization?

Pruning removes parts of the model. Quantization lowers the precision of what remains. They can be combined.

Does a pruned model always run faster?

Not automatically. Speedups need software and hardware that take advantage of the removed weights.

Sliding Window AttentionAn attention pattern where each token attends only to a fixed-size window of nearby tokens to reduce compute.Scaling LawAn empirical relationship showing how performance improves with model size, data, or compute.Distilled ModelA smaller model trained to imitate a larger model's behavior while using less compute at inference.BackpropagationThe core training algorithm that updates model weights by propagating prediction errors backward through the network.ComputeThe processing resources required to train and run models, often measured in FLOPS or GPU hours.GradientA vector showing how much each parameter should change to reduce loss.

Learn more in our free guides

Attention Rollout and Head Pruning Structured Pruning and Layer Dropping Model Pruning

Test yourself

Attention Rollout and Head Pruning Quiz Structured Pruning and Layer Dropping Quiz Model Pruning Quiz