Question

A data science team has spent three weeks running a workload on a GPU cluster where a neural network repeatedly processes a large labeled dataset, adjusting its internal weights after each pass to minimize prediction errors. A separate team now wants to take the resulting model and deploy it so a customer-facing app can return predictions on individual new inputs in real time. Which statement best characterizes these two phases?

Accepted Answer

The first phase is training, which is compute-intensive and updates model weights; the second phase is inference, which uses the fixed trained model to generate predictions on new data.

Answer

The first phase is inference because it repeatedly processes data; the second phase is training because it serves live users.

Answer

Both phases are training, since any use of a GPU with a neural network involves updating weights.

Answer

The first phase is fine-tuning and the second phase is quantization, which are the only two stages in a model lifecycle.