Snowflake Cortex Functions¶

COMPLETE¶

Given a prompt, the instruction-following COMPLETE function generates a response using your choice of the llama2-70b-chat or llama2-7b-chat model.

SNOWFLAKE.CORTEX.COMPLETE(model, prompt) -- returns string

• model: A string containing the name of the model to be used, either "llama2-70b-chat" or "llama2-7b-chat".

• prompt: A string containing the prompt to be used to begin generating the response. In an interactive chat scenario, this would be something the user typed; here, it is a plain-English description of what you want.

For example, to generate a single response:

SELECT SNOWFLAKE.CORTEX.COMPLETE('llama2-7b-chat', 'What are large language models?');

To generate a response from each of a table (in this example, content is a column from the reviews table):

SELECT SNOWFLAKE.CORTEX.COMPLETE(
    'llama2-7b-chat',
        CONCAT('Critique this review in bullet points: <review>', content, '</review>')
) FROM reviews LIMIT 10;

As shown in this example, you can use tagging in the prompt to control the kind of response generated. See A guide to prompting LLaMA 2 for tips.

SUMMARIZE¶

The SUMMARIZE function returns a summary of the given English text.

SNOWFLAKE.CORTEX.SUMMARIZE(text) -- returns string

• text: The text to be summarized.

To generate a summary for each row of a table (in this example, review_content is a column from the reviews table):

SELECT SNOWFLAKE.CORTEX.SUMMARIZE(review_content) FROM reviews LIMIT 10;

EXTRACT_ANSWER¶

The EXTRACT_ANSWER function extracts an answer to a question given in a text document.

SNOWFLAKE.CORTEX.EXTRACT_ANSWER(from_text, question) -- returns array

• from_text: The text that may contain the answer to the question. The text may be a plain-English document or a string representation of a semi-structured data object.

• question: The plain-English question to be answered. Be as specific as you can.

The result of this function is an array of answers, each an SQL object containing the following keys.

• answer: The answer extracted by the model.

• confidence: A floating-point value indicating the confidence level of the model in this answer.

• id: A unique identifier for the answer.

Answers are ordered with the highest confidence first. It is possible for the answers array to be empty if no answer can be found, or for even the best answer to have a low confidence score.

To extract an answer from each row of a table (in this example, review_content is a column from the reviews table):

SELECT SNOWFLAKE.CORTEX.EXTRACT_ANSWER(review_content,
    'What dishes does this review mention?')
FROM reviews LIMIT 10;

SENTIMENT¶

The SENTIMENT function returns sentiment as a score between -1 to 1 (with -1 being the most negative and 1 the most positive, with 0 neutral) for the given English-language input text.

SNOWFLAKE.CORTEX.SENTIMENT(text) -- returns float

• text: The text in which sentiment is to be detected.

For example, to score the sentiment of each row of a table (in this example, review_content is a column from the reviews table):

SELECT SNOWFLAKE.CORTEX.SENTIMENT(review_content) FROM reviews LIMIT 10;

TRANSLATE¶

The TRANSLATE function translates text from the indicated or detected source language to a target language.

SNOWFLAKE.CORTEX.TRANSLATE(text: string, from_language: enum, to_language: enum) → string

• text: The text to be translated.

• from_language: A language code from the list of supported languages indicating the original language of the text. If you pass an empty string instead of a language code, the TRANSLATE function attempts to detect the source language.

• to_language: A language code from the the list of supported languages indicating the language to translate the text into.

To translate each row of a table (in this example, review_content is a column from the reviews table):

SELECT SNOWFLAKE.CORTEX.TRANSLATE(review_content, 'en', 'de') FROM reviews LIMIT 10;

EMBED_TEXT¶

This function returns the vector embeddings for a given English-language text. This vector can be used with the vector comparison functions to determine the semantic similarity of two documents.

SNOWFLAKE.CORTEX.EMBED_TEXT(model, text) -- returns VECTOR

• model: The language model to be used to generate the embeddings. Currently this must be 〈e5-base-v2〉.

• text: The text for which an embedding should be calculated.

Model Restrictions¶

Models used by Snowflake Cortex have limitations on size as described in the table below. Sizes are given in tokens. Tokens generally correspond to words, but not all tokens are words, so the number of words corresponding to a limit is slightly less than the number of tokens. Inputs that exceed the limit are truncated to the permitted size.

^* Both input and output count toward the token limit, and by default output is limited to 512 tokens. You can change this using the optional max_tokens parameter.

Cost Considerations¶

During the initial phase of this preview, usage of Snowflake Cortex is free but limited. The per-account daily limits for each function are described in the table below.

After the initial phase (duration to be determined), using Snowflake Cortex will incur charges based on usage. The end of free usage will be announced well in advance along with full pricing details.

Upcoming Enhancements¶

We intend to deliver the following features and others in a future version of this preview. This intention does not, however, constitute a commitment to deliver any feature on any particular schedule, or at all.

• Python API, allowing the Cortex LLMs to be used from the preferred programming language of machine learning.

• REST API, allowing the Cortex LLMs to be used from any programming language that can make a RESTful Web services call.

• A text embedding function, allowing text documents to be compared for semantic similarity (see Vector Similarity).

Legal Notices¶

Cortex LLM Functions are powered by machine learning technology, including Meta’s LLaMA 2. The foundation LLaMA 2 model is licensed under the LLaMA 2 Community License and Copyright (c) Meta Platforms, Inc. All Rights Reserved. Your use of any LLM Functions based on the LLama 2 model is subject to Meta’s Acceptable Use Policy.

Machine learning technology and results provided may be inaccurate, inappropriate, or biased. Decisions based on machine learning outputs, including those built into automatic pipelines, should have human oversight and review processes to ensure model-generated content is accurate.

LLM function queries will be treated as any other SQL query and may be considered metadata.