Glossary term
Glossary term
Infrastructure and Serving
A printed circuit board (PCB) with multiple TPU chips, high bandwidth network interfaces, and system cooling hardware.
Created for this library
An ML platform team allocates TPU devices to training jobs based on memory and compute requirements.
A research engineer profiles model performance across TPU device generations to choose the most cost-effective option.
An ML platform team monitors TPU device health and rotates jobs off faulty devices automatically.
Definition source: Google for Developers Machine Learning Glossary | Creative Commons Attribution 4.0 License