AI4BharatRomanSetu

RomanSetu Romanized CPT + SFT

RomanSetu Romanized CPT + SFT is a distinct instruction-tuned checkpoint in the RomanSetu study.

Research release

The training order defines this research checkpoint

Romanized continual pretraining precedes Romanized instruction tuning.RomanSetu training configurations and published checkpoints

Keep native and Romanized evaluations separate

Practical guidance. A model's answer quality and its token efficiency are different measurements. Compare the script representation the model was trained to use, then inspect semantic accuracy.

Specifications & access

RomanSetu Romanized CPT + SFT specifications
DetailValue and conditions
StatusResearch release
AccessDownloadable
Model identifiersromansetu-cpt-roman-sft-roman
TermsLlama 2 license.RomanSetu Romanized CPT + SFT official model card
LanguagesEnglish, Hindi, Marathi, Gujarati, Tamil and Malayalam.RomanSetu Romanized CPT + SFT official model card

Related models

  • RomanSetu research series with five checkpoints of continued pretraining on Romanized Indian-language text.

    Launch date not confirmedResearch releaseDownloadable