Skip to content

Build the exact data slices each training run needs

Group curated records into reusable SFT, evaluation, and safety sets, then export the selected group in the format your stack expects.

Training groups interactive preview

How the workflow fits together

Groups preserve the intent behind a data slice. Keep record counts and review tags visible, select the target group, and export without reconstructing filters later.

  1. Define the slice

    Collect records around a training, evaluation, or safety objective.

  2. Review the composition

    Check record counts, subgroups, and curation tags before export.

  3. Export for the stack

    Choose a training-friendly format and download the selected group.

Reuse approved selections

Keep stable groups instead of rebuilding the same filters for every run.

Separate training purposes

Maintain distinct SFT, eval, and red-team data with clear intent.

Reduce export mistakes

Export the reviewed group rather than an ambiguous working dataset.

Prepare a cleaner handoff to training

Join the waitlist for early access to reusable training groups and exports.

Training groups · dropoutt