Use built-in platform models - PolarDB - Alibaba Cloud Documentation Center

Row-oriented AI includes three built-in platform models that you can deploy and call directly from SQL, without external ML infrastructure. This guide walks you through deploying a model, creating a SQL function, and running inference queries.

Function	Model name	File name	Return type	When to use
polarchat	builtin_polarchat	#ailib#_builtin_polarchat.so	STRING	Open-ended question answering over your data using LLMs.
polarzixun	builtin_polarzixun	#ailib#_builtin_polarzixun.so	STRING	Answers that draw from a knowledge base or document corpus, using retrieval-augmented LLMs.
qwen	builtin_qwen	#ailib#_builtin_qwen.so	STRING	General-purpose text generation and comprehension tasks, based on the Tongyi Qianwen model.

Prerequisites

Before you begin, ensure that you have:

Step 1: Deploy the built-in platform model

Connect to a database and run the following statement to deploy the model:

/*polar4ai*/ DEPLOY MODEL builtin_model_name WITH (mode = 'in_db');

Replace builtin_model_name with the model name from the table above (for example, builtin_polarchat). Set mode to in_db.

It may take a while to deploy the model. Run SHOW MODEL to check the status:

/*polar4ai*/ SHOW MODEL builtin_model_name;

Wait until the status changes to serving before proceeding.

Step 2: Create a function

Use the user-defined function (UDF) .so file generated in Step 1 to create a SQL function:

CREATE FUNCTION function_name RETURNS return_value SONAME "soname";

Parameter	Description
`function_name`	The predefined name of the built-in function. Must match the function name for the model deployed in Step 1.
`return_value`	The return type. Supported values: `REAL`, `STRING`, `INTEGER`.
`soname`	The `.so` file name. Must match the file name for the model deployed in Step 1.

For example, to create a function for builtin_polarchat:

CREATE FUNCTION polarchat RETURNS STRING SONAME "#ailib#_builtin_polarchat.so";

To verify the function was created, query the system table:

SELECT * FROM mysql.func;

Grant permissions to a standard account

To create a function, use a privileged account or grant a standard account the necessary permissions on the mysql.func system table.

Run the following statements using the privileged account:

GRANT INSERT ON mysql.* TO 'normal'@'localhost';
GRANT DELETE ON mysql.* TO 'normal'@'localhost';
FLUSH PRIVILEGES;

Step 3: Call the function for model inference

Call the function using a SELECT statement. Two calling patterns are supported:

Query a table — pass one or more column values as input:

SELECT function_name(feature1, feature2) FROM predict_table;

Pass a string directly — use a literal string as input:

SELECT function_name("content");

Parameter	Description
`function_name`	The function name created in Step 2.
`feature1`, `feature2`	Column names from the table used for inference.
`predict_table`	The table used for inference.
`content`	A literal string passed directly as input.

For example, to run interactive Q&A using polarchat against a support_tickets table:

SELECT polarchat(description) FROM support_tickets LIMIT 10;