Preview

This product is subject to the "Pre-GA Offerings Terms" in the General Service Terms section of theService Specific Terms. Pre-GA products are available "as is" and might have limited support. For more information, see thelaunch stage descriptions.

This page lists parameters for different functions provided by thegoogle_ml_integration extension to register and manage model endpoints. The page also lists the secrets that you can manage by using model endpoint management. To use AI models inproduction environments, seeBuild generative AI applications usingCloud SQL.

Models

Use this reference to understand parameters for functions that let you managemodel endpoints.

google_ml.create_model()

The following shows how to call thegoogle_ml.create_model() SQL function usedto register model endpoint metadata:

CALLgoogle_ml.create_model(model_id=>'MODEL_ID',model_request_url=>'REQUEST_URL',model_provider=>'PROVIDER_ID',model_type=>'MODEL_TYPE',model_qualified_name=>'MODEL_QUALIFIED_NAME',model_auth_type=>'AUTH_TYPE',model_auth_id=>'AUTH_ID',generate_headers_fn=>'GENERATE_HEADER_FUNCTION',model_in_transform_fn=>'INPUT_TRANSFORM_FUNCTION',model_out_transform_fn=>'OUTPUT_TRANSFORM_FUNCTION');

google_ml.alter_model()

The following shows how to call thegoogle_ml.alter_model() SQL function usedto update model endpoint metadata:

CALLgoogle_ml.alter_model(model_id=>'MODEL_ID',model_request_url=>'REQUEST_URL',model_provider=>'PROVIDER_ID',model_type=>'MODEL_TYPE',model_qualified_name=>'MODEL_QUALIFIED_NAME',model_auth_type=>'AUTH_TYPE',model_auth_id=>'AUTH_ID',generate_headers_fn=>'GENERATE_HEADER_FUNCTION',model_in_transform_fn=>'INPUT_TRANSFORM_FUNCTION',model_out_transform_fn=>'OUTPUT_TRANSFORM_FUNCTION');

google_ml.drop_model()

The following shows how to call thegoogle_ml.drop_model() SQL function usedto drop a model endpoint:

CALLgoogle_ml.drop_model('MODEL_ID');

google_ml.list_model()

The following shows how to call thegoogle_ml.list_model() SQL function usedto list model endpoint information:

SELECTgoogle_ml.list_model('MODEL_ID');

google_ml.model_info_view

The following shows how to call thegoogle_ml.model_info_view view that isused to list model endpoint information for all model endpoints:

SELECT*FROMgoogle_ml.model_info_view;

Secrets

Use this reference to understand parameters for functions that let you managesecrets.

google_ml.create_sm_secret()

The following shows how to call thegoogle_ml.create_sm_secret() SQL functionused to add the secret created in Secret Manager:

CALLgoogle_ml.create_sm_secret(secret_id=>'SECRET_ID',secret_path=>'projects/PROJECT_ID/secrets/SECRET_MANAGER_SECRET_ID/versions/VERSION_NUMBER');

google_ml.alter_sm_secret()

The following shows how to call thegoogle_ml.alter_sm_secret() SQL functionused to update secret information:

CALLgoogle_ml.alter_sm_secret(secret_id=>'SECRET_ID',secret_path=>'projects/PROJECT_ID/secrets/SECRET_MANAGER_SECRET_ID/versions/VERSION_NUMBER');

google_ml.drop_sm_secret()

The following shows how to call thegoogle_ml.drop_sm_secret() SQL functionused to drop a secret:

CALLgoogle_ml.drop_sm_secret('SECRET_ID');

Prediction functions

Use this reference to understand parameters for functions that let you generateembeddings or invoke predictions.

google_ml.embedding()

The following shows how to generate embeddings:

SELECTgoogle_ml.embedding(model_id=>'MODEL_ID',contents=>'CONTENT');

For examples of SQL queries to generate text embeddings, seeExamples.

google_ml.predict_row()

The following shows how to invoke predictions:

SELECTgoogle_ml.predict_row(model_id=>'MODEL_ID',request_body=>'REQUEST_BODY');

For examples of SQL queries to invoke predictions, seeExamples.

Transform functions

Use this reference to understand parameters for input and output transformfunctions.

Input transform function

The following shows the signature for the prediction function for text embeddingmodel endpoints:

CREATEORREPLACEFUNCTIONINPUT_TRANSFORM_FUNCTION(model_idVARCHAR(100),input_textTEXT)RETURNSJSON;

Output transform function

The following shows the signature for the prediction function for text embeddingmodel endpoints:

CREATEORREPLACEFUNCTIONOUTPUT_TRANSFORM_FUNCTION(model_idVARCHAR(100),response_jsonJSON)RETURNSreal[];

Transform functions example

To better understand how to create transform functions for your model endpoint,consider a custom-hosted text embedding model endpoint that requires JSON inputand output.

The following example cURL request creates embeddings based on the prompt andthe model endpoint:

curl-m100-XPOSThttps://cymbal.com/models/text/embeddings/v1\-H"Content-Type: application/json"-d'{"prompt": ["Cloud SQL Embeddings"]}'

The following example response is returned:

[[0.3522231-0.359320370.101560560.17734447-0.11606089-0.172660590.025093510.20305622-0.09787305-0.12154685-0.17313677-0.080754670.06821183-0.068965570.1171584-0.009315720.11875633-0.000774820.256049480.05193840.2034983-0.099526640.10347155-0.11935943-0.17872004-0.08706985-0.07056875-0.059293530.4177883-0.143817260.079349260.313682940.125432820.10758053-0.30210832-0.029510150.3908268-0.030910590.05302926-0.00114946-0.162337770.1117468-0.13159040.13947351-0.29569918-0.12330773-0.04354299-0.180689130.144455480.19481727]]

Based on this input and response, you can infer the following:

• The model expects JSON input through theprompt field. This field accepts anarray of inputs. As thegoogle_ml.embedding() function is a row levelfunction, it expects one text input at a time. Thus, you need to create aninput transform function that builds an array with a single element.

• The response from the model is an array of embeddings, one for each promptinput to the model. As thegoogle_ml.embedding() function is a row levelfunction, it returns a single input at a time. Thus, you need to create anoutput transform function that can be used to extract the embedding from thearray.

The following example shows the input and output transform functions that isused for this model endpoint when it is registered with model endpoint management:

CREATEORREPLACEFUNCTIONcymbal_text_input_transform(model_idVARCHAR(100),input_textTEXT)RETURNSJSONLANGUAGEplpgsqlAS$$DECLAREtransformed_inputJSON;model_qualified_nameTEXT;BEGINSELECTjson_build_object('prompt',json_build_array(input_text))::JSONINTOtransformed_input;RETURNtransformed_input;END;$$;

CREATEORREPLACEFUNCTIONcymbal_text_output_transform(model_idVARCHAR(100),response_jsonJSON)RETURNSREAL[]LANGUAGEplpgsqlAS$$DECLAREtransformed_outputREAL[];BEGINSELECTARRAY(SELECTjson_array_elements_text(response_json->0))INTOtransformed_output;RETURNtransformed_output;END;$$;

HTTP header generation function

The following shows signature for the header generation function that can beused with thegoogle_ml.embedding() prediction function when registering othertext embedding model endpoints.

CREATEORREPLACEFUNCTIONGENERATE_HEADERS(model_idVARCHAR(100),input_textTEXT)RETURNSJSON;

For thegoogle_ml.predict_row() prediction function, the signature is asfollows:

CREATEORREPLACEFUNCTIONGENERATE_HEADERS(model_idTEXT,inputJSON)RETURNSJSON;

Header generation function example

To better understand how to create a function that generates output in JSON keyvalue pairs that are used as HTTP headers, consider a custom-hosted textembedding model endpoint.

The following example cURL request passes theversion HTTP header, which isused by the model endpoint:

curl-m100-XPOSThttps://cymbal.com/models/text/embeddings/v1\-H"Content-Type: application/json"\-H"version: 2024-01-01"\-d'{"prompt": ["Cloud SQL Embeddings"]}'

The model expects text input through theversion field and returns the versionvalue in JSON format. The following example shows the header generation functionthat is used for this text embedding model endpoint when it is registered with modelendpoint management:

CREATEORREPLACEFUNCTIONheader_gen_fn(model_idVARCHAR(100),input_textTEXT)RETURNSJSONLANGUAGEplpgsqlAS$$BEGINRETURNjson_build_object('version','2024-01-01')::JSON;END;$$;

Request URL generation

Use the request URL generation function to infer the request URLs for the modelendpoints with built-in support. The following shows the signature for thisfunction:

CREATEORREPLACEFUNCTIONGENERATE_REQUEST_URL(providergoogle_ml.model_provider,model_typegoogle_ml.MODEL_TYPE,model_qualified_nameVARCHAR(100),model_regionVARCHAR(100)DEFAULTNULL)