Glossary term
Glossary term
Training and Fine-Tuning
When one number in your model becomes a NaN during training, which causes many or all other numbers in your model to eventually become a NaN.
NaN is an abbreviation for Not a Number.
Created for this library
An ML engineer adds NaN detection in the training loop to catch a NaN trap before it propagates across all parameters.
A research team disables certain unstable operations in mixed precision to prevent NaN traps in long training runs.
An ML platform team triggers automatic alerts whenever a NaN trap is detected in production training jobs.
Definition source: Google for Developers Machine Learning Glossary | Creative Commons Attribution 4.0 License