AI4BharatIndicBERT
IndicBERT v2 Shared Script
This 278M-parameter IndicBERT v2 release uses a distinct training configuration: Indic scripts are converted to Devanagari before masked-language training to increase lexical sharing.
The objective distinguishes this checkpoint
The model card links separate weights for the MLM, Samanantar, back-translation and shared-script experiments. These are training variants, not different service tiers.IndicBERT v2 Shared Script official model card
Control the downstream comparison
Practical guidance. Keep the script-conversion step in the evaluation. Comparing raw native-script input with a model trained on converted text would test the wrong pipeline.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | IndicBERTv2-SS |
| License | MIT.IndicBERT v2 Shared Script official model card |
| Evaluation | IndicXTREME; the release supports downstream classification, entity extraction, question answering and retrieval research.IndicBERT v2 Shared Script official model card |
Related models
- IndicBERT v2 Masked Language Model
AI4Bharat ยท IndicBERT
Multilingual IndicBERT v2 encoder trained with the masked-language-model objective.
Launch date not confirmedDownloadable