Snowflake Arctic-embed-xs
| Property | Value |
|---|---|
| Parameter Count | 22.6M |
| Embedding Dimension | 384 |
| Architecture | Based on all-MiniLM-L6-v2 |
| License | Apache-2.0 |
| Paper | Technical Report |
What is snowflake-arctic-embed-xs?
Snowflake Arctic-embed-xs is a compact yet powerful text embedding model that achieves state-of-the-art retrieval performance in its size class. Despite having only 22.6M parameters, it delivers performance comparable to models with 100M parameters, making it ideal for production deployments with strict latency and cost requirements.
Implementation Details
The model is built on the all-MiniLM-L6-v2 architecture and trained through a multi-stage pipeline optimized for retrieval performance. It generates 384-dimensional embeddings and achieves an impressive MTEB Retrieval Score (NDCG@10) of 50.15, significantly outperforming other models in its class like GIST-all-MiniLM-L6-v2 (45.12) and gte-tiny (44.92).
- Multi-stage training pipeline with 400M samples
- Optimized for retrieval tasks with hard negative mining
- 384-dimensional dense embeddings
- Support for both sentence-transformers and HuggingFace implementations
Core Capabilities
- Efficient text embedding generation
- Strong retrieval performance (50.15 NDCG@10)
- Low computational requirements
- Support for query-document similarity scoring
- Easy integration with popular frameworks
Frequently Asked Questions
Q: What makes this model unique?
Its ability to achieve near-100M-parameter model performance with only 22.6M parameters, making it extremely efficient for production deployments.
Q: What are the recommended use cases?
The model is ideal for text retrieval tasks, semantic search, and document similarity applications where computational efficiency is crucial.
