Lightning-AI/pytorch-lightning
Auf GitHub ansehenenable loading `universal checkpointing` checkpoint in `DeepSpeedStrategy`
Open
#20.065 geöffnet am 9. Juli 2024
featurehelp wantedstrategy: deepspeed
Repository-Metriken
- Stars
- (26.687 Stars)
- PR-Merge-Metriken
- (Durchschn. Merge 9T 15h) (3 gemergte PRs in 30 T)
Beschreibung
Description & Motivation
After I trained a model in some numbers of gpus, say, 8 gpus for a while, It's difficult to load the checkpoint to 16 gpus with optimizer and model states unchanged. The deepspeed has developed the universal checkpointing strategy to solve this problem, but I didn't see the pytorch-lightning has this feature.
Pitch
I want the pytorch-lightning could support this feature
Alternatives
try to add universal_checkpoint as a param of DeepSpeedStrategy and modify the class refering to https://www.deepspeed.ai/tutorials/universal-checkpointing/
Additional context
No response
cc @borda @awaelchli