The answer
Separate test data helps estimate performance on examples the model has not optimized against. Reusing test answers during development makes that estimate less trustworthy.
What does it mean to train a neural network?
Why are training and test data separated?
More FAQs in What does it mean to train a neural network?
Usually, answering a question uses existing parameters. Updating those parameters requires a separate training procedure, although a product may independently store conversation history or collect feedback.
An epoch is one complete pass through a training dataset. Large language model training may also be described by tokens processed or optimization steps.
Still have questions?
Browse all our FAQs or reach out to our support team
Want to upskill yourself?
Your learning library
Small steps. Big possibilities.
Find your next course. Build something great.