Skip to main content

Model Deployment

Our models can be deployed in any of the following ways:

  • Docker container
  • Server (FastAPI)
  • Model Registry (MLFlow)
  • Cognaize Platform

Docker Container​

The model can be deployed as a Docker container. The container can be by either using the build.sh in the repository or by running the following command:


docker build -t <image_name> .
docker run <image_name> python driver.py <arguments>

Server​

The model can be deployed as a server using FastAPI. To do so, run the following can be run:


uvicorn server:app --reload

The model will be ready to be used in http://localhost:8000. And the documentation will be available in http://localhost:8000/docs.

Endpoints​

Run model by giving document as input​

This functionality allows you to run the model locally, by giving a document as input.

Endpoint: /run/predict/genie
Method: POST

  • Body:

    {
        "document_json": "str",
        "data_path": "str"
    }

  • Body validations:

    • document_json: Not blank
    • data_path: Not blank

  • Response Body

    {
        "status": "str",
        "error": "str",
        "result": "dict"
    }

Run model, and digest to cognaize platform​

This functionality allows you to run the model locally, while reading the data from the platform and digesting the results back to the platform.

Endpoint: /run/genie/
Method: POST

  • Body:

    {
        "task_id": "str",
        "token": "str",
        "url": "str"
    }

  • Body validations:

    • task_id: Not blank
    • token: Not blank
    • url: Not blank

  • Response Body

    {
        "task_id": "str"
    }

Run base model​

This functionality allows you to run the model locally, by giving a document as input.

Endpoint: /run/predict/genie
Method: POST

  • Body:

    {
        "input_type": "str",
        "input": "str | UploadedFile"
    }

  • Body validations:

    • input_type: Not blank
    • input: Not blank

  • Response Body

    {
        "status": "str",
        "error": "str",
        "result": "dict"
    }