Promptsy is a Python library designed for managing and organizing prompts for language models in a structured way. It provides a convenient method to store, retrieve, and format prompts using YAML files.
- Save prompts to YAML files with a specified name and directory structure
- Load prompts from YAML files by name
- Format prompts with dynamic values using keyword arguments
- Automatically create directories for prompts if they don't exist
- List all available prompts in the specified base directory
- Colorized error messages for better visibility
- Enhance prompts using OpenAI's language model
- Generate few-shot examples for prompts using the
FewShotPromptGenerator, allowing for improved context and response generation based on specified expected outputs.
To install Promptsy, you can use pip:
pip install promptsyBefore using the PromptEnhancer, you need to set up your OpenAI API key. You can do this by setting the OPENAI_API_KEY environment variable. Here’s how to set it:
set OPENAI_API_KEY=your_api_key_hereexport OPENAI_API_KEY=your_api_key_hereReplace your_api_key_here with your actual OpenAI API key.
Here's a basic example of how to use Promptsy:
from promptsy.prompt import Prompt
# Create an instance of Prompt
prompt = Prompt(
name="hello_world",
description="An example prompt",
template="Hello, {name}!"
)
# Save the prompt using the Prompt's save method
prompt.save() # Não é necessário passar o PromptManager
# Load the prompt using the Prompt's load method
loaded_prompt = Prompt.load("hello_world") # Não é necessário passar o PromptManager
print(loaded_prompt) # Output: hello_world: An example prompt
print(loaded_prompt.format(name="Taylor Swift")) # Output: Hello, Taylor Swift!The PromptManager class is responsible for managing the storage and retrieval of prompts. It provides methods for saving prompts to YAML files, loading prompts from YAML files, and listing all available prompts.
manager = PromptManager(base_directory='prompts')base_directory(optional): The base directory where the prompts will be stored. Defaults to'prompts'.
prompt.save(name) # O método save agora é chamado diretamente no objeto Promptloaded_prompt = Prompt.load(name) # O método load agora é chamado diretamente na classe Promptname(str): The name of the prompt to load.- Returns: The loaded prompt data.
prompts = manager.list_prompts()- Returns: A list of prompt names.
The PromptEnhancer class allows you to enhance prompts using OpenAI's language model. It generates improved versions of prompts based on the original template.
enhancer = PromptEnhancer(api_key='your_api_key_here')api_key(str, optional): OpenAI API key. If None, it tries to retrieve from the environment variable.model_name(str, optional): The name of the model to be used (default: "gpt-4o-mini").
enhanced_prompt = enhancer.enhance_prompt(prompt)prompt(Prompt): A Prompt object containing the original prompt template.- Returns: A new Prompt object with the enhanced template.
Here’s an example of how to use the PromptEnhancer with the prompt_toddlers_story_time:
from promptsy.prompt import Prompt
from promptsy.prompt_manager import PromptManager
from promptsy.prompt_enhancer import PromptEnhancer
# Create an instance of PromptManager
manager = PromptManager()
# Create an instance of Prompt for toddlers story time
prompt_toddlers_story_time = Prompt(
name="toddlers_story_time",
description="Write a bedtime story for toddlers",
template="You are a bedtime story teller for toddlers. Write a story for a toddler about a {animal} that goes on an adventure to {place}."
)
# Save the original prompt
prompt_toddlers_story_time.save()
# Create an instance of PromptEnhancer
enhancer = PromptEnhancer()
# Enhance the prompt
enhanced_prompt = enhancer.enhance_prompt(prompt_toddlers_story_time)
print("Enhanced Prompt:")
print(enhanced_prompt.template)The enhanced prompts are automatically saved in the enhanced_prompts directory.
Here’s an example of how to use the FewShotPromptGenerator to generate few-shot examples for a sentiment analysis prompt:
from promptsy.auto_few_shot_generator import FewShotPromptGenerator
from promptsy.prompt import Prompt
# Create an instance of Prompt for sentiment analysis
sentiment_analysis_prompt = Prompt(
name="sentiment_analysis",
description="Generate a sentiment analysis prompt",
template="You are a sentiment analysis prompt generator. Generate a prompt for sentiment analysis for the following text: {text}"
)
# Create an instance of FewShotPromptGenerator
few_shot_generator = FewShotPromptGenerator(model_name="gpt-4o-mini")
# Generate few-shot examples using the sentiment analysis prompt
num_examples = 3
expected_outputs = ["negative", "positive", "neutral"]
sentiment_analysis_prompt_with_examples = few_shot_generator.generate_examples(
prompt_initial=sentiment_analysis_prompt,
num_examples=num_examples,
expected_outputs=expected_outputs,
return_examples=False
)
print("Formatted Prompt with Few-Shot Examples:")
print(sentiment_analysis_prompt_with_examples.template)
# Optionally, you can call the OpenAI API with the generated prompt
final_prompt = sentiment_analysis_prompt_with_examples.format(text="I am so happy today, but I am tired!")
print(final_prompt)Contributions are welcome! If you find any issues or have suggestions for improvements, please open an issue or submit a pull request on the GitHub repository.
This project is licensed under the MIT License.