Skip to content

Add support for cross-encoder models #75

Closed
@peakji

Description

@peakji

Feature request

Add support for cross-encoder models, such as BAAI/bge-reranker-base and various models from the Sentence Transformers - Cross-Encoders project.

A new API endpoint could also be added, for example:

curl 127.0.0.1:8080/predict \
    -X POST \
    -d '{"inputs": ["What is Deep Learning?", "Deep Learning is ..."]}' \
    -H 'Content-Type: application/json'

Motivation

In vector search, we often employ a cross-encoder model for reranking the top-k results, which typically yields better performance compared to using embedding models alone.

Your contribution

Cross-encoder models share similar model structures with embedding models, suggesting that some optimizations could potentially be reused.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions