Weight Streaming — AI Dictionary

Running inference by loading each layer's weights from host memory or storage into accelerator memory just before that layer executes, rather than holding the whole model on-device at once.