Class LlamaIndexQueryPipelineAgent (1.86.0)
Stay organized with collections Save and categorize content based on your preferences.

LlamaIndexQueryPipelineAgent(
    model: str,
    *,
    system_instruction: typing.Optional[str] = None,
    prompt: typing.Optional[QueryComponent] = None,
    model_kwargs: typing.Optional[typing.Mapping[str, typing.Any]] = None,
    model_builder: typing.Optional[typing.Callable[[...], FunctionCallingLLM]] = None,
    retriever_kwargs: typing.Optional[typing.Mapping[str, typing.Any]] = None,
    retriever_builder: typing.Optional[typing.Callable[[...], QueryComponent]] = None,
    response_synthesizer_kwargs: typing.Optional[
        typing.Mapping[str, typing.Any]
    ] = None,
    response_synthesizer_builder: typing.Optional[
        typing.Callable[[...], QueryComponent]
    ] = None,
    runnable_kwargs: typing.Optional[typing.Mapping[str, typing.Any]] = None,
    runnable_builder: typing.Optional[typing.Callable[[...], QueryPipeline]] = None,
    enable_tracing: bool = False
)

A LlamaIndex Query Pipeline Agent.

This agent uses a query pipeline for LLAIndex, including prompt, model, retrieval and summarization steps. More details can be found in https://docs.llamaindex.ai/en/stable/module_guides/querying/pipeline/.

Methods

LlamaIndexQueryPipelineAgent

LlamaIndexQueryPipelineAgent(
    model: str,
    *,
    system_instruction: typing.Optional[str] = None,
    prompt: typing.Optional[QueryComponent] = None,
    model_kwargs: typing.Optional[typing.Mapping[str, typing.Any]] = None,
    model_builder: typing.Optional[typing.Callable[[...], FunctionCallingLLM]] = None,
    retriever_kwargs: typing.Optional[typing.Mapping[str, typing.Any]] = None,
    retriever_builder: typing.Optional[typing.Callable[[...], QueryComponent]] = None,
    response_synthesizer_kwargs: typing.Optional[
        typing.Mapping[str, typing.Any]
    ] = None,
    response_synthesizer_builder: typing.Optional[
        typing.Callable[[...], QueryComponent]
    ] = None,
    runnable_kwargs: typing.Optional[typing.Mapping[str, typing.Any]] = None,
    runnable_builder: typing.Optional[typing.Callable[[...], QueryPipeline]] = None,
    enable_tracing: bool = False
)

Initializes the LlamaIndexQueryPipelineAgent.

Under-the-hood, assuming .set_up() is called, this will correspond to

# model_builder
model = model_builder(model_name, project, location, model_kwargs)

# runnable_builder
runnable = runnable_builder(
    prompt=prompt,
    model=model,
    retriever=retriever_builder(model, retriever_kwargs),
    response_synthesizer=response_synthesizer_builder(
        model, response_synthesizer_kwargs
    ),
    runnable_kwargs=runnable_kwargs,
)

When everything is based on their default values, this corresponds to a query pipeline Prompt - Model:

# Default Model Builder
model = google_genai.GoogleGenAI(
    model=model_name,
    vertexai_config={
        "project": initializer.global_config.project,
        "location": initializer.global_config.location,
    },
)

# Default Prompt Builder
prompt = prompts.ChatPromptTemplate(
    message_templates=[
        types.ChatMessage(
            role=types.MessageRole.USER,
            content="{input}",
        ),
    ],
)

# Default Runnable Builder
runnable = QueryPipeline(
    modules = {
        "prompt": prompt,
        "model": model,
    },
)
pipeline.add_link("prompt", "model")

When system_instruction is specified, the prompt will be updated to include the system instruction.

# Updated Prompt Builder
prompt = prompts.ChatPromptTemplate(
    message_templates=[
        types.ChatMessage(
            role=types.MessageRole.SYSTEM,
            content=system_instruction,
        ),
        types.ChatMessage(
            role=types.MessageRole.USER,
            content="{input}",
        ),
    ],
)

When all inputs are specified, this corresponds to a query pipeline Prompt - Model - Retriever - Summarizer:

runnable = QueryPipeline(
    modules = {
        "prompt": prompt,
        "model": model,
        "retriever": retriever_builder(retriever_kwargs),
        "response_synthesizer": response_synthesizer_builder(
            response_synthesizer_kwargs
        ),
    },
)
pipeline.add_link("prompt", "model")
pipeline.add_link("model", "retriever")
pipeline.add_link("model", "response_synthesizer", dest_key="query_str")
pipeline.add_link("retriever", "response_synthesizer", dest_key="nodes")

Parameters
Name	Description
`model`	`str` The name of the model (e.g. "gemini-1.0-pro").
`system_instruction`	`str` Optional. The system instruction to use for the agent.
`prompt`	`llama_index.core.base.query_pipeline.query.QUERY_COMPONENT_TYPE` Optional. The prompt template for the model.
`model_kwargs`	`Mapping[str, Any]` Optional. Keyword arguments for the model constructor of the google_genai.GoogleGenAI. An example of a model_kwargs is: python { # api_key (string): The API key for the GoogleGenAI model. # The API can also be fetched from the GOOGLE_API_KEY # environment variable. If `vertexai_config` is provided, # the API key is ignored. "api_key": "your_api_key", # temperature (float): Sampling temperature, it controls the # degree of randomness in token selection. If not provided, # the default temperature is 0.1. "temperature": 0.1, # context_window (int): The context window of the model. # If not provided, the default context window is 200000. "context_window": 200000, # max_tokens (int): Token limit determines the maximum # amount of text output from one prompt. If not provided, # the default max_tokens is 256. "max_tokens": 256, # is_function_calling_model (bool): Whether the model is a # function calling model. If not provided, the default # is_function_calling_model is True. "is_function_calling_model": True, }
`model_builder`	`Callable` Optional. Callable that returns a language model.
`retriever_kwargs`	`Mapping[str, Any]` Optional. Keyword arguments for the retriever constructor.
`retriever_builder`	`Callable` Optional. Callable that returns a retriever object.
`response_synthesizer_kwargs`	`Mapping[str, Any]` Optional. Keyword arguments for the response synthesizer constructor.
`response_synthesizer_builder`	`Callable` Optional. Callable that returns a response_synthesizer object.
`runnable_kwargs`	`Mapping[str, Any]` Optional. Keyword arguments for the runnable constructor.
`runnable_builder`	`Callable` Optional. Callable that returns a runnable (query pipeline).
`enable_tracing`	`bool` Optional. Whether to enable tracing. Defaults to False.

clone

clone() -> (
    vertexai.preview.reasoning_engines.templates.llama_index.LlamaIndexQueryPipelineAgent
)

Returns a clone of the LlamaIndexQueryPipelineAgent.

query

query(
    input: typing.Union[str, typing.Mapping[str, typing.Any]], **kwargs: typing.Any
) -> typing.Union[
    typing.Dict[str, typing.Any], typing.Sequence[typing.Dict[str, typing.Any]]
]

Queries the Agent with the given input and config.

Parameter
Name	Description
`input`	`Union[str, Mapping[str, Any]]` Required. The input to be passed to the Agent.

set_up

set_up()

Sets up the agent for execution of queries at runtime.

It initializes the model, connects it with the prompt template, retriever and response_synthesizer.

This method should not be called for an object that being passed to the ReasoningEngine service for deployment, as it initializes clients that can not be serialized.

Class LlamaIndexQueryPipelineAgent (1.86.0) Stay organized with collections Save and categorize content based on your preferences.