Web App
NEWBuild Reproducible Training Dataset Mixtures
Sample multiple sources into a documented dataset mixture.
Screenshot
Inspect the interface
Description
What is this software?
Sample multiple sources into a documented dataset mixture.
Features
- Combine source datasets using declared weights, record caps, token budgets and a reproducible seed.
- Choose sampling with or without replacement and retain a source-row ledger for every selection.
Sample multiple sources into a documented dataset mixture.
Features
- Combine source datasets using declared weights, record caps, token budgets and a reproducible seed.
- Choose sampling with or without replacement and retain a source-row ledger for every selection.

