/
Embedding Models used by Pragatix

Embedding Models used by Pragatix

 

Pragatix supports both on-prem and cloud embedding models. For on-prem deployments, intfloat/multilingual-e5-large is a supported option, and Alibaba-NLP/gte-Qwen2-1.5B-instruct is available as an alternate model. Cloud options include text-embedding-3-small and Cohere Embed v4 through AWS Bedrock.

Embedding Models

Pragatix supports multiple embedding models. These are grouped into two deployment categories:

  • On-Prem models — embedding models hosted in the customer environment

  • Cloud models — embedding models accessed as managed external services

Customers may choose either option. Even when all Pragatix software and customer data are installed on-premises, a cloud embedding model can still be used.

On-Prem Models

Primary option: intfloat/multilingual-e5-large

intfloat/multilingual-e5-large is one of the supported on-prem embedding models in Pragatix.

The E5 family originates from Microsoft research. In this context, “Microsoft E5” refers to the E5 embedding model family, not to any Microsoft E5 license. Its research and release lineage is reflected in Microsoft Research publications and Microsoft’s UNILM GitHub repository, which includes the E5 release.

  • Model page: https://huggingface.co/intfloat/multilingual-e5-large

  • Microsoft UNILM GitHub: https://github.com/microsoft/unilm/tree/master/e5

  • Additional internal documentation: https://agatsoftware.atlassian.net/wiki/spaces/VA/pages/4861853698/E5-Large+Embedding+Model

The E5 model is also available as a cloud service through OpenRouter.

Alternate option: Alibaba-NLP/gte-Qwen2-1.5B-instruct

As an alternative to E5, Pragatix also supports gte-Qwen2-1.5B-instruct as an on-prem embedding model.

  • Model page: https://huggingface.co/Alibaba-NLP/gte-Qwen2-1.5B-instruct

Cloud Models

OpenAI text-embedding-3-small

Supported as a cloud embedding option.

  • Model page: https://developers.openai.com/api/docs/models/text-embedding-3-small

AWS Bedrock / Cohere Embed v4

Supported as a cloud embedding option.

  • Model page: https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-embed-v4.html