AI4BharatIndicBERT
IndicBERT
IndicBERT is the original ALBERT-based multilingual encoder. It was pretrained on about 9 billion tokens across English and 11 Indian languages.
Restricted access
A compact encoder predates the later BERT and Gemma releases
Its evaluation covers classification, question answering, entity recognition and cross-language retrieval. This is a pretrained representation model, not an instruction-following chat model.IndicBERT official model card
Add a task-specific training step
Practical guidance. Use the encoder to investigate language features or train a downstream task. A raw hidden-state vector is not automatically a calibrated search embedding.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Restricted access |
| Access | Downloadable |
| Model identifiers | indic-bert |
| License | MIT; the current Hugging Face download is gated.IndicBERT official model card |
| Languages | Assamese, Bengali, English, Gujarati, Hindi, Kannada, Malayalam, Marathi, Odia, Punjabi, Tamil and Telugu.IndicBERT 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