Granite Embedding Models

2025-02-27

AI Summary

This work addresses the longstanding trade-off between efficiency and effectiveness in multilingual information retrieval (IR) embedding models. We introduce the Granite family of embedding models—comprising 12-layer dense/sparse base models and 6-layer distilled lightweight variants—supporting both English and multilingual retrieval. Our approach innovatively integrates retrieval-oriented pretraining, contrastive fine-tuning, knowledge distillation, and model ensembling, jointly optimized on enterprise-grade, high-quality, compliant data to balance accuracy, inference efficiency, and regulatory requirements. On IBM’s internal retrieval benchmarks, Granite significantly outperforms open-source models of comparable size; it achieves state-of-the-art (SOTA) results on standard IR benchmarks including MSMARCO and BEIR. All Granite models are released under the Apache 2.0 license, providing an efficient, production-ready, and trustworthy open-source foundation for industrial multilingual IR systems.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Language Grounding & Multi-modal NLP
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

We introduce the Granite Embedding models, a family of encoder-based embedding models designed for retrieval tasks, spanning dense-retrieval and sparse retrieval architectures, with both English and Multilingual capabilities. This report provides the technical details of training these highly effective 12 layer embedding models, along with their efficient 6 layer distilled counterparts. Extensive evaluations show that the models, developed with techniques like retrieval oriented pretraining, contrastive finetuning, knowledge distillation, and model merging significantly outperform publicly available models of similar sizes on both internal IBM retrieval and search tasks, and have equivalent performance on widely used information retrieval benchmarks, while being trained on high-quality data suitable for enterprise use. We publicly release all our Granite Embedding models under the Apache 2.0 license, allowing both research and commercial use at https://huggingface.co/collections/ibm-granite.

Problem

• Develop embedding models for retrieval tasks
• Compare performance on retrieval and search benchmarks
• Release models for research and commercial use

Innovation

• encoder-based embedding models
• retrieval oriented pretraining techniques
• multilingual dense-sparse retrieval architectures