v0.3.0

What's New

• Dataset diversity metrics (vocabulary richness, n-gram diversity, topic entropy)
• Token budget calculator with cost awareness
• 55+ new tests

v0.2.0

What's Changed in v0.2.0

• add automatic dataset ratio optimization
• add streaming dataset support
• add YAML and JSON config format for mix recipes

Full Changelog: v0.1.0...v0.2.0

datamix v0.1.0

What's New

First stable release of datamix — Dataset mixing and curriculum optimizer

Highlights

• Full Python API + CLI
• 135 passing tests
• Python 3.9–3.13 support
• Zero or minimal dependencies
• Apache 2.0 license

Install

pip install datamix