SEA-LION-ModernBERT-Embedding-300M
SEA-LION-ModernBERT-Embedding-300M is the smaller tuned ModernBERT encoder in AI Singapore's embedding collection. It uses a custom Gemma 3 SentencePiece tokenizer and a 262K vocabulary to represent regional text.
How this release works
The model is contrastively trained on 245 million text pairs, then instruction-tuned on 13 million pairs. The card lists eleven languages, including Chinese and English alongside nine regional languages. It supplies a Sentence Transformers example for semantic similarity.SEA-LION-ModernBERT-Embedding-300M official model card
How to assess it for your work
Practical guidance. Start with this encoder when building a regional search index and compare it with the 600M version on the same questions. Keep passage boundaries and ranking rules fixed during the comparison. Measure whether the correct evidence appears among the first results, rather than judging only the numerical similarity score.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | aisingapore/SEA-LION-ModernBERT-Embedding-300M |
| Architecture | ModernBERT-base encoderSEA-LION-ModernBERT-Embedding-300M official model card |
| Model size class | 300MSEA-LION-ModernBERT-Embedding-300M official model card |
| Context stated by the card | 8K tokensSEA-LION-ModernBERT-Embedding-300M official model card |
| License declared by the model card | MITSEA-LION-ModernBERT-Embedding-300M official model card |
Related models
- SEA-LION-E5-Embedding-600M
AI Singapore · SEA-LION Embedding
A multilingual E5 text encoder that produces 1,024-dimensional embeddings for eleven regional languages.
Launch date not confirmedDownloadable - SEA-LION-ModernBERT-Embedding-600M
AI Singapore · SEA-LION Embedding
A 600M ModernBERT text encoder for regional retrieval and similarity tasks with a custom Gemma 3 tokenizer.
Launch date not confirmedAPI · Downloadable