Memory efficient MAML - Githubissues

cbfinn / maml

Code for "Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks"

MIT License

2.55k stars 604 forks source link

Memory efficient MAML #72

Open dbaranchuk opened 4 years ago

dbaranchuk commented 4 years ago

Hi,

We would like to share the application of the gradient checkpointing technique to MAML. This allows performing way (~10-100x) more MAML steps with the same GPU memory budget. Maybe someone dealing with MAML will find it useful for their projects.

PyTorch implementation

Example:

Any comments and proposals are highly appreciated.