Why can training use more memory than inference?
Training commonly keeps activations, gradients, and optimizer state in addition to model parameters. Inference usually avoids much of that extra storage and computation.
Explore This Question
Search for additional explanations using these tools:
More FAQs in What is a forward pass in neural-network training?
A normal forward pass calculates outputs using current weights. Learned parameter updates happen through the optimizer, although some training-mode layers may update separate running statistics.
Parameters are learned values reused across inputs. Activations are intermediate results of processing a particular input and can change even when parameters stay fixed.
Still have questions?
Browse all our FAQs or reach out to our support team