Text Embeddings
Models
DS1 currently provides two embedding models with the following specifications:
| Model Name | Dimension | Context Length | Tokenizer | Description |
|---|---|---|---|---|
| DS1-EN-V1 | 512 (L2 normalized) | 512 tokens | 30k WordPiece | High-performance English text retrieval model |
| DS1-MULTILINGUAL-V1 | 256 (L2 normalized) | 200,000 max tokens/request | Multilingual tokenizer | Multilingual retrieval model for cross-language semantic search |
Tokenization
Tokenizer configuration differs by model:
| Model Name | Vocabulary Size | Tokenizer Config | Special Tokens |
|---|---|---|---|
| DS1-EN-V1 | 30,000 | model_max_length: 512 | [PAD], [UNK], [CLS] |
| DS1-MULTILINGUAL-V1 | 250,002 | model_max_length: 200K | BOS, CLS, EOS, MASK, PAD, SEP, UNK |
Modality
DS1 is a text-only embedding model family optimized for semantic search and retrieval.
Runtime Request Limits
The following limits are runtime API/container limits and are not intrinsic model capabilities.
- Use up to 32 texts per request.
- Keep payloads compact to avoid endpoint memory pressure on smaller instances.
- For large corpora, process in chunks of 16 to 32 texts and retry failed chunks.
Want to know about additional models? Check out our FAQ.