zandervdm/flashrank-api repository overview

⁠FlashRank-API

FlashRank-API is a FastAPI-based REST service that provides text reranking capabilities using the FlashRank library. It allows you to rerank a list of passages based on their relevance to a given query using various BERT-based models.

⁠Features

• Fast and efficient text reranking
• Support for multiple BERT models
• Configurable maximum sequence length
• Built-in model caching
• Health check endpoint
• RESTful API interface

⁠Quick Start

docker pull flashrank-api:latest
docker run -p 8000:8000 flashrank-api:latest

The API will be available at http://localhost:8000.

⁠API Endpoints

⁠POST /rerank

Reranks a list of passages based on their relevance to a query.

⁠Request Body

{
  "query": "your search query",
  "passages": [
    {
      "id": "1",
      "text": "passage content",
      "meta": {
        "optional_key": "optional_value"
      }
    }
  ],
  "max_length": 128,
  "model": "ms-marco-MultiBERT-L-12"
}

Parameters:

• query (required): The search query to rank passages against
• passages (required): Array of passages to rank
  • id: Unique identifier for the passage (any type)
  • text: The passage content
  • meta: Optional metadata object
• max_length (optional): Maximum sequence length for the model (default: 128)
• model (optional): Name of the BERT model to use (default: "ms-marco-MultiBERT-L-12")

⁠Response

{
  "results": [
    {
      "id": "1",
      "score": 0.95,
      "text": "passage content",
      "meta": {
        "optional_key": "optional_value"
      }
    }
  ]
}

⁠GET /health

Health check endpoint to verify service status.

⁠Response

{
  "status": "healthy"
}

⁠Technical Details

• Base image: python:3.9-slim
• Working directory: /app
• Exposed port: 8000
• Model cache directory: /opt/flashrank_cache
• Server: uvicorn

⁠License

This project is licensed under the MIT License - see the LICENSE file for details.