What is Gradient Checkpointing?

Gradient checkpointing is a memory-saving training technique that trades compute for memory by recomputing intermediate activations during the backward pass rather than storing them from the forward pass. The platform strengthens