End-to-end user journeys for generative AI models
You can use remote or hosted generative AI models in BigQuery ML to integrate advanced AI capabilities directly into BigQuery. Exploring these end-to-end user journeys lets you discover the specific SQL statements and functions required to create, evaluate, and generate data from Gemini models, Cloud AI services, and custom open models. BigQuery ML offers the following types of remote models:
- Fine-tuned Google Gemini models
- Google, partner, and open models as a service
- Google text embedding models as a service
- Self-deployed open models
- Cloud AI services
- Custom models deployed to Gemini Enterprise Agent Platform
Remote or hosted model user journeys
The following table describes the statements and functions you can use to create, evaluate, and generate data from remote or hosted models:
| Model category | Model type | Model creation | Evaluation | Inference | Tutorials |
|---|---|---|---|---|---|
| Generative AI remote models | Remote model over a Gemini text generation model1 |
CREATE MODEL
|
ML.EVALUATE |
||
| Remote model over a partner text generation model |
CREATE MODEL
|
ML.EVALUATE |
AI.GENERATE_TEXT
|
N/A | |
| Remote model over an open text generation model3 |
CREATE MODEL
|
ML.EVALUATE |
AI.GENERATE_TEXT
|
Generate text with Gemma and public data | |
| Remote model over a Google embedding generation model |
CREATE MODEL
|
N/A |
AI.GENERATE_EMBEDDING
|
||
| Remote model over an open embedding generation model3 |
CREATE MODEL
|
N/A |
AI.GENERATE_EMBEDDING
|
Generate text embeddings by using an open model and the AI.GENERATE_EMBEDDING function
|
|
| Cloud AI remote models | Remote model over the Cloud Vision API |
CREATE MODEL
|
N/A |
ML.ANNOTATE_IMAGE
|
Annotate images |
| Remote model over the Cloud Translation API |
CREATE MODEL
|
N/A | Translate text | ||
| Remote model over the Cloud Natural Language API |
CREATE MODEL
|
N/A |
ML.UNDERSTAND_TEXT
|
Understand text | |
| Remote model over the Document AI API |
CREATE MODEL
|
N/A |
ML.PROCESS_DOCUMENT
|
||
| Remote model over the Speech-to-Text API |
CREATE MODEL
|
N/A |
ML.TRANSCRIBE
|
Transcribe audio files | |
| Remote model over a custom model deployed to Gemini Enterprise Agent Platform | Remote model over a custom model deployed to Gemini Enterprise Agent Platform | CREATE MODEL |
ML.EVALUATE |
ML.PREDICT |
Make predictions with a custom model |
1 Some Gemini models support supervised tuning.
2 This function calls a hosted Gemini model, and
doesn't require you to create a model separately using the CREATE MODEL
statement.
3 You can automatically deploy an open model when you create the
BigQuery ML remote model by specifying the model's Hugging Face
or Agent Platform Model Garden ID. BigQuery
manages the Agent Platform resources of open models deployed in this
way, and lets you interact with those Agent Platform resources by
using the BigQuery ML ALTER MODEL and DROP MODEL
statements. It also lets you configure automatic undeployment of the model.
For more information, see
Automatically deployed models.
4 This function calls the Cloud Translation API directly, and
doesn't require you to create a model separately by using the CREATE MODEL
statement.