RomanSetu CPT Native Script
This RomanSetu series studies continual pretraining with text in its native script. The publisher provides 100m, 200m, 300m, 400m and 500m checkpoints.
Research release
Five training checkpoints belong to one controlled series
The numbered checkpoints are training stages, not model parameter counts. The article groups their identifiers for discovery. The corresponding instruction-tuned releases are separate.RomanSetu training configurations and published checkpoints
Compare stages with a fixed text representation
Practical guidance. Keep the input script and downstream tuning procedure unchanged when comparing this series. Otherwise the result combines a training-stage change with a preprocessing change.
Published training checkpoints
romansetu-cpt-native-100m; romansetu-cpt-native-200m; romansetu-cpt-native-300m; romansetu-cpt-native-400m; romansetu-cpt-native-500m.RomanSetu native 100m published checkpointRomanSetu native 200m published checkpointRomanSetu native 300m published checkpointRomanSetu native 400m published checkpointRomanSetu CPT Native Script official model card
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | romansetu-cpt-native-100mromansetu-cpt-native-200mromansetu-cpt-native-300mromansetu-cpt-native-400mromansetu-cpt-native-500m |
| Base family | Llama 2; model-card license is llama2.RomanSetu CPT Native Script official model card |
| Listed languages | English, Hindi, Marathi, Gujarati, Tamil and Malayalam.RomanSetu CPT Native Script official model card |
Related models
- RomanSetu CPT Romanized
AI4Bharat ยท RomanSetu
RomanSetu research series with five checkpoints of continued pretraining on Romanized Indian-language text.
Launch date not confirmedResearch releaseDownloadable