Models
Every model we have released. Each one is small enough to run on an ordinary computer, and each ships with its weights, code and a write-up of how it was trained and where it fails. How they compare is on the benchmarks page.
Netra OCR 1
Reads Khmer text in photos, scans and documents
- Parameters
- 19M
- License
- MIT
- Trained on
- 1.3M images
Netra MT 1
Translates between English and Khmer, in both directions
- Parameters
- 77M
- License
- MIT
- Trained on
- 220M tokens
AkaraAlpha
Type Khmer in Latin letters, the way you'd text it, and get it back in Khmer script
- Parameters
- 78.7K
- CER
- 15.07%
- License
- MIT
- Trained on
- 28,569 samples
Sota
Transcribes Khmer and English speech, at a fifth of its teacher's size
- Parameters
- 73M / 204M
- CER
- 1.77%
- License
- Apache-2.0
- Trained on
- 1,537 hours