Foundation Model Stage Overview

📘

Note

This article refers to BaseModel accessed via Docker container. Please refer to Snowflake Native App section if you are using BaseModel as SF GUI application.

The training of the foundation model is a critical step in BaseModel as described here.
It lays the groundwork for creating downstream models tailored to specific needs of your business.
It is the trigger to capitalization of the benefits delivered by the solution.

During this stage, BaseModel:

• connects to data sources, automatically detecting the types of columns and relationships among them in order to apply the best possible feature extractors,
• applies our proprietary algorithms, such as Cleora _and _emde, in order to create behavioral profiles reflecting complex relationships between entities,
• trains deep learning model based on our optimized neural network architecture to build the robust foundation for subsequent tasks,
• saves the model outputs as an input for downstream models.

For a foundation model to be built, the user needs to:

1. Define data sources using YAML configuration file
2. (optionally) Modify default training parameters using the same file
3. Initiate training with a Python function or via command line