Skip to main content
The Curator Viewer is a hosted web page where you can look through your data. When it is on, Curator uploads each response as it arrives, so you can watch the dataset grow during a run.

​
Turn on the viewer

Set the CURATOR_VIEWER environment variable before you run Curator.
Then run any Curator script. Curator prints a line like this next to the progress display.
Each run gets its own link. You can also read the link from the viewer_url attribute of the CuratorResponse.
If you do not set an API key, anyone with the link can see the dataset. Set a Bespoke Labs API key to keep your datasets private.

​
Use a Bespoke Labs API key

When you set an API key, Curator links your datasets to your Bespoke Labs account. You can then do these things.
  • Keep your datasets private.
  • See all the datasets you have made.
  • Share datasets with other people.
  • See what your data generation cost over time.
1

Create an API key

Sign in to the Bespoke Labs console and create a key.
2

Set the environment variables

Curator now streams every dataset to the viewer and links it to your account.
3

Find your datasets

The Datasets page lists the datasets made with your keys and the ones others shared with you. The Cost report page shows what you spent on data generation over a period.

​
Upload an existing dataset

Use push_to_viewer to upload a Hugging Face Dataset that you already have. You can also pass the ID of a dataset on the Hugging Face Hub. The function returns the viewer link, and it works without CURATOR_VIEWER.

​
Download a dataset from the viewer

Use load_dataset with the dataset ID, which is the last part of the viewer link. It returns a Hugging Face Dataset and caches it in the Curator cache directory.
Curator does not send your API key with this request, so use it for datasets that anyone with the link can open.
The local curator-viewer command is retired. Use the hosted viewer instead.