mixed precision training

A technique where certain parts of the network use lower precision (like float16) to speed up computation and reduce memory usage while maintaining model quality.