Hugging Face has released Granite Embedding Multilingual R2, a suite of open-source multilingual embedding models. The release includes a 97M-parameter compact model that leads in retrieval quality among open models under 100M parameters and a larger 311M-parameter model that ranks second among open models under 500M parameters. Both models support over 200 languages, handle a 32K token context window, and are trained on code retrieval across nine programming languages, all under the Apache 2.0 license. AI
IMPACT Enhances multilingual capabilities for retrieval-augmented generation and cross-lingual search with open-source models.
RANK_REASON Release of open-source embedding models with benchmark results.
- Apache 2.0
- Granite Embedding Multilingual R2
- Haystack
- Hugging Face
- LangChain
- LlamaIndex
- Milvus
- ModernBERT
- MTEB Multilingual Retrieval
- ONNX
- OpenVINO
- IBM
AI-generated summary · Google Gemini · from 3 sources. How we write summaries →
