Overview
The purpose of this article is to provide a high-level overview of Snowflake’s AI/ML offerings – Snowflake Cortex. This is intended to provide a consolidated overview of the different types of functionality and applications that fall under the Cortex umbrella. In future articles, we will dive deeper into the technical functionality.
What is Snowflake Cortex?
At a high-level, Snowflake Cortex is their managed AI & ML offering which enables users to analyze data and build generative AI applications using fully managed LLMs, vector search and fully managed text-to-SQL services. All with minimal ML experience. Snowflake Cortex is broken down into 3 core functionalities:
- LLM functions
- ML functions
- Features Powered by Cortex
Let’s break down each section.
LLM functionality
Snowflake Cortex gives you instant access to industry-leading large language models (LLMs) trained by researchers at companies like Mistral, Reka, Meta, and Google. This also includes Snowflake Arctic, their own open enterprise-grade model.
Since these LLMs are fully hosted and managed by Snowflake, using them requires no setup. Your data stays within Snowflake, giving you the performance, scalability, and governance you expect.
Cortex LLM functionalities are broken down into 3 sections:
- Complete function
- Task-specific function
- Helper functions
Complete Function
The COMPLETE function takes a prompt (and the model used for it) from a user and returns a response, similar to other LLMs. Though, true to Snowflake, this is all done within SQL commands.
Task-Specific Functions
These types of functions are leveraged automate routine-tasks (e.g. summaries, translations, sentiment, etc.), that don’t require any customization. Specific functions include:
Helper Functions
These managed functions reduce cases of failures when running other LLM functions. For example, getting the count of tokens in an input prompt to ensure the call doesn’t exceed a model limit. Helped functions include:
ML functionality
These powerful analysis functions give you automated predictions and insights into your data using machine learning. ML functionality is broken down into 2 sections:
- Time-series
- Non-time series
Time-Series Functions
These features train a machine learning model on your time-series data to determine how a specified metric varies over time and relative to other features of your data. The model then provides insights or predictions based on the trends detected in the data. Specific functions include:
- Forecasting – predicts future metric values from past trends in time-series data.
- Anomaly Detection – flags metric values that differ from typical expectations.
Non Time-Series Functions
These features don’t require time series data and include:
- Classification – sort rows into two or more classes based on their most predictive features.
- Top Insights – helps you find dimensions and values that affect the metric in surprising ways.
Features Powered by Cortex
Within the Cortex umbrella, there are applications that are powered by the underlying functions. This includes:
Summary
The article provides an overview of Snowflake Cortex, a managed AI/ML offering that allows users to analyze data and create generative AI applications with minimal machine learning expertise. It highlights three core functionalities: large language model (LLM) functions, machine learning (ML) functions, and various applications powered by Cortex, including Document AI and Universal Search. We look forward to guiding our clients on the next phase of their journey to leverage their new fully managed LLMs and automated ML capabilities without hindering performance and data governance.
Comments are closed