In simple words: Training teaches; validation guides choices; testing checks the finished model.
Three data sets
Training data teaches the model. Validation data helps choose settings and when to stop. Test data gives a final check on new examples after choices are made. Keep the test set separate so its result is meaningful.
Hyperparameters
A hyperparameter is a setting chosen for the learning process, such as learning rate, batch size, number of layers, or a maximum number of epochs. Model weights are different: training learns them from data.
See the flow
train→tune on validation→final test
Simple example
With 1,000 old records, you might use 700 to train, 150 to validate, and 150 to test. The exact split depends on the project.
Remember: Training teaches; validation guides choices; testing checks the finished model.