Wandb.link_model and metadata

Inside the Model Registry app each model version has an associated Metadata tab in the UI where the user can see information related to the specific model. It’s a useful feature, but it has issues:

  • The API reference for link_model doesn’t say anything about how to populate the Metadata tab when linking a model. The guide also didn’t make this really clear.
  • From what I’ve gathered the only way to programmatically populate Metadata is to include a .yaml file in the directory pointed to by the path argument. This is really inconvenient - can the link_model function be updated accept an optional metadata kwarg of type dict?

Hello, This a reply from our support bot designed to assist you with your Weights & Biases related queries. To reach a human please reply to this message.

The Weights & Biases Model Registry provides a comprehensive way to manage the lifecycle of your models, from training to production. Within the Model Registry, each model version has an associated Metadata tab in the UI, which displays information related to that specific model version. However, based on the provided documentation, there isn’t a direct mention of how to programmatically populate the Metadata tab through the link_model function or any other function with a metadata keyword argument.

The current method to include metadata for a model version involves adding a .yaml file in the directory that you’re pointing to with the path argument when you’re creating or linking a model artifact. This .yaml file should contain the metadata you wish to associate with the model version. When the model artifact is created or linked, the contents of the .yaml file are automatically recognized and displayed in the Metadata tab in the UI.

Your suggestion to update the link_model function (or a similar function within the W&B SDK) to accept an optional metadata keyword argument of type dict is insightful. This would indeed streamline the process of associating metadata with model versions by allowing users to specify metadata programmatically at the time of linking the model, rather than having to rely on a separate .yaml file.

As of the information available in the provided documentation, there isn’t a direct API call or function parameter that supports this functionality. For enhancements or feature requests like this, it’s best to reach out directly to the Weights & Biases support team or submit a feature request through their community forums or GitHub repository. The W&B team is very receptive to user feedback and continuously works on improving their platform based on user needs.

For now, to programmatically include metadata with your model versions, you would need to follow the workaround of including a .yaml file with the desired metadata in your model’s directory. If you’re looking for more detailed guidance or have further questions, I recommend contacting W&B support at support@wandb.com or exploring the community forums at https://wandb.me/community for more insights and updates on this feature.

To reach a human please reply to this message.

-WandBot :robot:

Hi Piotr,

We wanted to follow up with you regarding your support request as we have not heard back from you. Please let us know if we can be of further assistance or if your issue has been resolved.

Weights & Biases

Hi Piotr, since we have not heard back from you we are going to close this request. If you would like to re-open the conversation, please let us know!