microsoft / DeepSpeed

DeepSpeed is a deep learning optimization library that makes distributed training and inference easy, efficient, and effective.
https://www.deepspeed.ai/
Apache License 2.0
33.63k stars 3.95k forks source link

Fix memory leak from _hp_mapping #5643

Closed chiragjn closed 4 days ago

chiragjn commented 2 weeks ago

See #5496 I don't really know if this is a good solution

chiragjn commented 1 week ago

Apologies, I had made some mistakes in bf16 optimizer and couldn't get to fixing them sooner, I should have tested things locally 😅