Skip to main content

portia.execution_agents.utils.step_summarizer

StepSummarizer implementation.

The StepSummarizer can be used by agents to summarize the output of a given tool.

StepSummarizer Objects​

class StepSummarizer()

Class to summarize the output of a tool using llm.

This is used only on the tool output message.

Attributes:

  • summarizer_prompt ChatPromptTemplate - The prompt template used to generate the summary.
  • model GenerativeModel - The language model used for summarization.
  • summary_max_length int - The maximum length of the summary.
  • step Step - The step that produced the output.

__init__​

def __init__(config: Config,
             model: GenerativeModel,
             tool: Tool,
             step: Step,
             summary_max_length: int = 500) -> None

Initialize the model.

Arguments:

  • config Config - The configuration for the run.
  • model GenerativeModel - The language model used for summarization.
  • tool Tool - The tool used for summarization.
  • step Step - The step that produced the output.
  • summary_max_length int - The maximum length of the summary. Default is 500 characters.

invoke​

def invoke(state: MessagesState) -> dict[str, Any]

Invoke the model with the given message state.

This method processes the last message in the state, checks if it's a tool message with an output, and if so, generates a summary of the tool's output. The summary is then added to the artifact of the last message.

Arguments:

  • state MessagesState - The current state of the messages, which includes the output.

Returns:

dict[str, Any]: A dict containing the updated message state, including the summary.

Raises:

  • Exception - If an error occurs during the invocation of the summarizer model.