mlp-python-service-template

This repository is a project template for creating a new trainable Caila service.

Caila is a platform for hosting microservices based on ML models. It is a powerful tool that can cover every aspect of your solution’s lifecycle, from model training and QA to deployment and monitoring.

Create a new project from this template to start developing a service of your own!

Get started

Start by getting yourself acquainted with the contents of main.py. In terms of features, this is a simple “Hello World” service:

• Its fit method accepts json/texts dataset and "trains" the demo model.
• Its predict method returns parts of the dataset the model was trained by in fit.

The service relies on the Caila Python SDK to expose its functionality to the platform.

Build Docker image

To build the service locally, run ./build.sh in the project root. You need to have Docker Engine installed and running.

The build script will create a Docker image, push it to the public Caila Docker registry, and print the image URL to the console:

--------------------------------------------------
Docker image: docker-pub.caila.io/caila-public/mlp-trainable-service-xxxxxxxxxxxxxxxx:main
--------------------------------------------------

You will need this URL to configure your service in Caila.

⚠ The public Caila Docker registry has a limited storage time and is intended for educational purposes only. Do not use it for production.

Create Caila service

1. Sign in to Caila or sign up for a new account.
2. Go to My space and select Images in the sidebar.

   🛈 If you don’t see this tab, go to My space → Services, select Create service, and submit a request for access. Our customer support team will get back to you shortly.

3. Select Create image. Provide the image name and the URL you got from the build script.
4. On the image description page, select Create service. Provide the service name, mark model as Fittable and select json/texts in FitDatasetType. Leave the other settings at their defaults.
5. You should now see your service in the Services tab. Go to its page and click on Fit. To upload the dataset select Upload, choose type json/texts, fill the dataset name and attach a file with the following content: {"texts": ["It's the first sentence", "and it's the second one"]}. Then fill a name for the new fitted service and put an empty JSON ({}) into Fit configuration. Click on Fit and Go to a new service.
6. Wait for the fit to be completed (the status indicator should turn from yellow to green, and you should see SUCCESS status).
7. You can check events related to the training process in Events history menu.
8. Go to the Testing and try sending a request with a JSON body like {"texts": [0]} or {"texts": [0, 1]}.

Your service is up and running if you can see the output with the sentences from the dataset file you uploaded.

If you would like to learn more about Caila, check out our official documentation.

License

This project is licensed under Apache License 2.0.