Prompt

Overview

The Prompt class is an extension of the Prompt class from the promptfile library. It provides additional functionality for managing prompts, including response formats, few-shot examples, and cost tracking.

Attributes

AttributeTypeDescription
modelstringThe name of the language model to use for this prompt.
messagesList[ChatCompletionMessageParam]A list of messages that make up the prompt. Each message is a dictionary containing role and content.
response_formatOptional[ResponseFormat]Specifies the desired format for the model’s response. Can be a dictionary or a custom pydantic BaseModel.
fewshotOptional[List[DatasetItem]]A list of few-shot examples to be injected into the prompt template.
temperatureOptional[float]Controls the randomness of the model’s output. Lower values make the output more deterministic.
nameOptional[string]An optional name for the prompt, useful for identification and cost tracking.
inputs_descOptional[Dict[str, str]]A dictionary describing the expected input parameters for the prompt.
outputs_descOptional[Dict[str, str]]A dictionary describing the expected output format from the model.

Methods

__call__

Asynchronously calls the language model with the configured prompt.

Parameters:

  • lm_config: Optional dictionary for additional language model configuration.
  • num_retries: Number of retries for API calls (default: 3).
  • **kwargs: Additional keyword arguments for formatting the prompt.

Returns: The response from the language model, either as a string or a dictionary.

load

Loads a Prompt object from a string content.

Parameters:

  • content: The string content to load the prompt from.

Returns: A new Prompt object.

load_file

Loads a Prompt object from a file.

Parameters:

  • file_path: The path to the file to load the prompt from.

Returns: A new Prompt object.

format

Formats the prompt with given keyword arguments.

Parameters:

  • **kwargs: Keyword arguments to format the prompt with.

Returns: The formatted Prompt object.

reset_copy

Creates a reset copy of the Prompt object with empty few-shot examples.

Returns: A new Prompt object with reset few-shot examples.

dump

Dumps the Prompt object to a string representation.

Returns: A string representation of the Prompt object.

Usage

import textwrap
from pydantic import BaseModel
from ape.common import Prompt
 
class Response(BaseModel):
  reasoning: str
  answer: str
 
base_prompt = Prompt(
    messages=[
        {
            "role": "system",
            "content": "You are a helpful assistant."
        },
        {
            "role": "user",
            "content": "Answer the following question: {question}"
        }
    ],
    model="gpt-4o",
    response_format=Response
)
 
response: Response = await prompt(
  question="What is the meaning of life?"
)
ape-commonDatasetItem