Unlocking the Power of Snowflake Cortex for Generative AI

Snowflake Cortex offers seamless access to some of the most capable enterprise-grade models, including Vellama, Mistral, Reka, and the Cloud family of models. These powerful tools serve as the foundation for building advanced Generative AI applications with ease and precision.

LLM Functions: Simplified Building Blocks for Gen AI Apps

Snowflake Cortex provides three types of Large Language Model (LLM) functions:

1. Complete Function: A versatile, stateless function to handle general tasks.
2. Task-Specific Functions: Purpose-built for common tasks like translation, sentiment analysis, summarization, and text classification.
3. Helper Functions: Designed to assist in prototyping and deploying AI applications into production.

Task-Specific Functions

Task-specific functions simplify AI development by eliminating the need for manual prompts or model specification. Examples include:

• Cortex Translate: Enables text translation by passing the text or table column, source language, and target language. If the source language is unknown, it can auto-detect the language by leaving the source field empty. Cortex Translate supports 11 languages.
• ClassifyText: Categorizes free-form text into predefined categories, making text classification straightforward and efficient.

Helper Functions for Prototyping and Production

To ensure robust application development, Cortex provides two key helper functions:

1. Try Complete:
2. Count Tokens:

Cortex Access Control

To safeguard Cortex’s capabilities, access is restricted to users with the Cortex user role. This ensures the secure and controlled use of these advanced AI tools.

The Complete Function: Stateless and Configurable

The Complete function serves as a versatile building block for AI applications:

• Model Parameter: Specifies the model to use.
• Stateless Nature: Does not retain state between calls; previous prompts and responses must be included as part of the input to enable a stateful experience.
• Advanced Options: Includes parameters like max tokens, temperature, and top-p for fine-tuning model behaviour.