ChatFirebaseVertexAI class

Wrapper around Vertex AI for Firebase API (aka Gemini API).

Example:

final chatModel = ChatFirebaseVertexAI();
final messages = [
  ChatMessage.humanText('Tell me a joke.'),
];
final prompt = PromptValue.chat(messages);
final res = await chatModel.invoke(prompt);

• Vertex AI for Firebase

Setup

To use ChatFirebaseVertexAI you need to have:

• A Firebase project with Blaze pay-as-you-go pricing plan
• aiplatform.googleapis.com and firebaseml.googleapis.com APIs enabled
• Firebase SDK initialized in your app
• Recommended: Firebase App Check enabled

Available models

The following models are available:

• gemini-1.5-flash:
  • text / image / audio -> text model
  • Max input token: 1048576
  • Max output tokens: 8192
• gemini-1.5-pro:
  • text / image / audio -> text model
  • Max input token: 2097152
  • Max output tokens: 8192
• gemini-1.0-pro-vision:
  • text / image -> text model
  • Max input token: 16384
  • Max output tokens: 2048
• gemini-1.0-pro
  • text -> text model
  • Max input token: 32760
  • Max output tokens: 8192

Mind that this list may not be up-to-date. Refer to the documentation for the updated list.

Call options

You can configure the parameters that will be used when calling the chat completions API in several ways:

Default options:

Use the defaultOptions parameter to set the default options. These options will be used unless you override them when generating completions.

final chatModel = ChatFirebaseVertexAI(
  defaultOptions: ChatFirebaseVertexAIOptions(
    model: 'gemini-1.5-pro-preview',
    temperature: 0,
  ),
);

Call options:

You can override the default options when invoking the model:

final res = await chatModel.invoke(
  prompt,
  options: const ChatFirebaseVertexAIOptions(temperature: 1),
);

Bind:

You can also change the options in a Runnable pipeline using the bind method.

In this example, we are using two totally different models for each question:

final chatModel = ChatFirebaseVertexAI();
const outputParser = StringOutputParser();
final prompt1 = PromptTemplate.fromTemplate('How are you {name}?');
final prompt2 = PromptTemplate.fromTemplate('How old are you {name}?');
final chain = Runnable.fromMap({
  'q1': prompt1 | chatModel.bind(const ChatFirebaseVertexAIOptions(model: 'gemini-1.0-pro')) | outputParser,
  'q2': prompt2 | chatModel.bind(const ChatFirebaseVertexAIOptions(model: 'gemini-1.0-pro-vision')) | outputParser,
});
final res = await chain.invoke({'name': 'David'});

Tool calling

ChatFirebaseVertexAI supports tool calling.

Check the docs for more information on how to use tools.

Example:

const tool = ToolSpec(
  name: 'get_current_weather',
  description: 'Get the current weather in a given location',
  inputJsonSchema: {
    'type': 'object',
    'properties': {
      'location': {
        'type': 'string',
        'description': 'The city and state, e.g. San Francisco, CA',
      },
    },
    'required': ['location'],
  },
);
final chatModel = ChatFirebaseVertexAI(
  defaultOptions: ChatFirebaseVertexAIOptions(
    model: 'gemini-1.5-pro',
    temperature: 0,
    tools: [tool],
  ),
);
final res = await model.invoke(
  PromptValue.string('What’s the weather like in Boston and Madrid right now in celsius?'),
);