A city food-pantry inventory script loads a transformer and then crawls, even though the machine has a GPU the process never used. What is the framework-project fix?
Select an answer to reveal the explanation.
Short Explanation
The pantry script loaded a transformer and then crawled on CPU while a GPU sat idle. Move the model and its batch tensors onto that GPU. Do not rewrite attention as a kernel, do not lengthen the prompt, and do not file a cooling ticket.
Full Explanation
Associate device placement means sending the model and its batch to the GPU the machine already has. The intern does not rewrite attention as a CUDA kernel. A longer prompt on CPU makes the crawl worse, not better. A slow local script that never touched the GPU is an application bug, not a datacenter cooling ticket.