AI engineers can now measure how much to shrink their models without breaking them

Researchers created a new way to measure how efficient a shrunken AI model is. This helps engineers choose the right "size" for an AI without making it useless.

AI models often need to run on small devices like phones or cheap sensors. Engineers try to shrink these models, but it is hard to know how much to shrink them before they stop working. This new measurement tool tells them exactly when a model becomes too small for its job. It also shows that some common shrinking methods can completely break certain types of AI.