leap_mapping.yaml

Describes the purpose of the leap_mapping.yaml

When uploading a model and dataset to the Tensorleap server, the platform expects the user to connect UI nodes in the Network tab. This serves as a way to point to which loss, metrics & visualizers connect to which GT, input or predictions.

After creating these connections, the platform allow you to save these connections in the form of a yaml. Downloading this mapping into your local code-base would ensure that on following upload of the same architecture there would be no need to repeat the process of connecting the UI nodes.

An example of leap_mapping.yaml for the MNIST use-case could be found here.

Working with multiple mappings

In case you aim to use Tensorleap with multiple models you would need multiple leap_mapping.yaml files. It is advised to place them in a single place within the repo, and per model select the relevant model when uploading a new model to the platform by using the --leap-mapping flag and directing the CLI to the path of the relevant mapping.