Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Granite Embedding Models

Parul Awasthy, Aashka Trivedi, Yulong Li, Mihaela Bornea, David Cox, Abraham Daniels, Martin Franz, Gabe Goodhart, Bhavani Iyer, Vishwajeet Kumar, Luis Lastras, Scott McCarley, Rudra Murthy, Vignesh P, Sara Rosenthal, Salim Roukos, Jaydeep Sen, Sukriti Sharma, Avirup Sil, Kate Soule, Arafat Sultan, Radu Florian

TL;DR

Granite Embedding Models address enterprise retrieval needs by providing encoder-based embeddings that support dense and sparse representations in English and multilingual contexts. The approach combines retrieval-oriented pretraining, contrastive finetuning, knowledge distillation, and model merging to achieve strong performance on public benchmarks and IBM-specific tasks, while respecting licensing and data governance. The models span 30M and 125M dense English variants, 107M and 278M multilingual variants, and a 30M sparse English variant, with 6- and 12-layer configurations and 512-token inputs, enabling a range of latency-accuracy tradeoffs. Public release under Apache 2.0 facilitates broad deployment in research and enterprise settings.

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.

Granite Embedding Models

TL;DR

Granite Embedding Models address enterprise retrieval needs by providing encoder-based embeddings that support dense and sparse representations in English and multilingual contexts. The approach combines retrieval-oriented pretraining, contrastive finetuning, knowledge distillation, and model merging to achieve strong performance on public benchmarks and IBM-specific tasks, while respecting licensing and data governance. The models span 30M and 125M dense English variants, 107M and 278M multilingual variants, and a 30M sparse English variant, with 6- and 12-layer configurations and 512-token inputs, enabling a range of latency-accuracy tradeoffs. Public release under Apache 2.0 facilitates broad deployment in research and enterprise settings.

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.

Paper Structure

This paper contains 43 sections, 11 equations, 1 figure, 16 tables.

Table of Contents

  1. Introduction
  2. Granite Embedding Models
  3. Model Architecture
  4. Training Recipe
  5. Training Data
  6. Weakly Paired Data with In-Batch Negatives
  7. Annotated Data with Hard Negatives
  8. Synthetic Data
  9. Hard Negative Mining
  10. Data Sampling
  11. Dense Retrievers: English
  12. granite-embedding-125m-english
  13. Retrieval Oriented Pretraining
  14. Decoder based teacher model
  15. Distill from decoder based model
  16. ...and 28 more sections

Figures (1)

  • Figure 1: Average performance on the Granite embedding models (in blue) vs BGE, GTE, Snowflake, E5, and Nomic models on 5 QA and IR datasets: BEIR, ClapNQ, CoIR, RedHat, and UnifiedSearch (the last 2 are internal IBM datasets).