Log model checkpoints#

Uploading model checkpoints#

You can save model weights from any deep learning framework by using the upload() method.

Model checkpoints appear in the All metadata section. In the below example, they're logged under a field called my_model in the namespace model_checkpoints.

What you logWhat you get

Log PyTorch model weights

my_model = ...
torch.save(my_model, "my_model.pt")
run["model_checkpoints/my_model"].upload("model_checkpoints/my_model.pt")

Model checkpoints in all metadata

See example in Neptune

Tracking checkpoint metadata#

You can also upload the model checkpoints to S3-compatible storage and only track their metadata with Neptune.

For instructions, see Track artifacts.