Microsoft's 270M edge-ready multilingual embedder with 32K context.
Access model weights, configuration files, and documentation.
See which devices can run this model and at what quality level.
The smallest Harrier variant from Microsoft Research, a 268M-parameter Gemma 3-based decoder embedding model designed to run on CPU and edge devices. Trained with contrastive learning plus knowledge distillation from larger teachers, supports 94 languages and 32K-token contexts, with an MIT license enabling commercial use.
View on Hugging Face