Skip to main content

Types of services in Caila

Simple service​

A simple service performs one predict operation. It can either be called via the API if you are an external user, or run on your own servers if you are a service developer.

Service requiring deployment: prototype​

These services are provided as a template that you can run on your server. When deploying in your own account, you could modify the service settings and configure its scaling parameters.

Fittable service​

These are services that require training on user data. For example, an intent classifier (short texts) or LLM fine-tuning. As a result of training (or using the factory), a new derived service will be created in the user’s workspace.

Basic​

A basic service serves as the starting point for training. Basic services are located in the Catalogue.

Derived​

A derived service is a service obtained as a result of training. A derived service is created in the user’s account in My space.

Composite service​

A composite service calls to other services within its logic. Calls can occur during both predict and fit operations.

The simplest example of a composite service is the text-classifier-logreg-caila-roberta intent classifier. This classifier uses the BERT model from another service vectorizer-caila-roberta.

Application​

These are standalone web applications with their own interface. Typically, they are designed to showcase the capabilities of ML services, facilitate testing, or carry out specific practical tasks.