- Notifications
You must be signed in to change notification settings - Fork0
A simple library to document Pydantic models for structured LLM outputs using standard Python docstrings.
License
OpenAdaptAI/PydanticPrompt
Folders and files
| Name | Name | Last commit message | Last commit date | |
|---|---|---|---|---|
Repository files navigation
A simple library to documentPydantic models for structured LLM outputs using standard Python docstrings.
pip install pydantic-prompt
- Python 3.11 or higher
- Pydantic 2.0 or higher
- Document Pydantic model fields using standard Python docstrings
- Format field documentation for LLM prompts with a single method call
- Automatic type inference and display
- Optional validation rule documentation
- Warnings for undocumented fields
- Seamlessly embeds documentation in your prompts
frompydanticimportBaseModel,FieldfromtypingimportOptionalfrompydantic_promptimportprompt_schema@prompt_schemaclassPerson(BaseModel):name:str"""The person's full name"""age:int=Field(ge=0)"""Age in years"""email:Optional[str]=None"""Contact email address if available"""# Use in LLM promptprompt=f"""Please extract a Person object from this text according to the schema below:{Person.format_for_llm()}Input: Jane Smith, 28, works at Acme Corp."""
Theformat_for_llm() method produces output like:
Person:- name (str): The person's full name- age (int): Age in years- email (str, optional): Contact email address if availablefrompydanticimportBaseModelfrompydantic_promptimportprompt_schemafromtypingimportlist@prompt_schemaclassAddress(BaseModel):street:str"""Street name and number"""city:str"""City name"""postal_code:str"""Postal or ZIP code"""@prompt_schemaclassContact(BaseModel):name:str"""Full name of the contact"""addresses:list[Address]= []"""List of addresses associated with this contact"""# The format_for_llm method will nicely format nested structures tooprint(Contact.format_for_llm())
You can include validation rules in the output:
Person.format_for_llm(include_validation=True)
Output:
Person:- name (str): The person's full name- age (int): Age in years [Constraints: ge: 0]- email (str, optional): Contact email address if availableBy default, PydanticPrompt warns you when fields don't have docstrings:
@prompt_schemaclassPartiallyDocumented(BaseModel):name:str"""This field has a docstring"""age:int# Missing docstring will generate a warning
Output:
UserWarning: Field 'age' in PartiallyDocumented has no docstring. Add a docstring for better LLM prompts.To disable these warnings:
@prompt_schema(warn_undocumented=False)classSilentModel(BaseModel):name:str# No warning for missing docstring
frompydanticimportBaseModel,Fieldfrompydantic_promptimportprompt_schemafromtypingimportlist@prompt_schemaclassProductReview(BaseModel):product_name:str"""The exact name of the product being reviewed"""rating:int=Field(ge=1,le=5)"""Rating from 1 to 5 stars, where 5 is best"""pros:list[str]"""List of positive aspects mentioned about the product"""cons:list[str]"""List of negative aspects mentioned about the product"""summary:str"""A brief one-sentence summary of the overall review"""# In your LLM promptprompt=f"""Extract a structured product review from the following text.Return the data according to this schema:{ProductReview.format_for_llm(include_validation=True)}Review text:I recently purchased the UltraBook Pro X1 and I've been using it for about two weeks.The screen is absolutely gorgeous and the battery lasts all day which is fantastic.However, the keyboard feels a bit mushy and it runs hot when doing anything intensive.Overall, a solid laptop with a few minor issues that don't detract from the experience."""
PydanticPrompt uses Python's introspection capabilities to:
- Capture docstrings defined after field declarations
- Extract type information from type hints
- Identify optional fields
- Format everything in a clear, LLM-friendly format
The library is designed to be lightweight and focused on a single task: making it easy to document your data models for LLMs in a way that both developers and AI can understand.
- Consistency: Document your models once, use everywhere
- DRY Principle: No need to duplicate field descriptions between code and prompts
- Developer-Friendly: Uses standard Python docstrings
- LLM-Friendly: Produces clear, structured documentation that LLMs can easily understand
- Maintainability: When your models change, your prompt documentation automatically updates
Clone the repository:
git clone https://github.com/OpenAdaptAI/PydanticPrompt.gitcd PydanticPromptCreate and activate a virtual environment withuv:
# Create a virtual environmentuv venv# Activate the virtual environmentsource .venv/bin/activate# On Unix/macOS# or.venv\Scripts\activate# On Windows
Install the package in development mode:
# Install the packageuv pip install -e.
Install development dependencies:
# Install dev dependenciesuv pip install -e".[dev]"
If you see a warning about hardlinking when installing:
warning: Failed to hardlink files; falling back to full copy. This may lead to degraded performance.This is normal when the cache and target directories are on different filesystems. You can suppress this warning with:
export UV_LINK_MODE=copyMake sure your virtual environment is activated, then run:
# Run all testspytest# Run with verbose outputpytest -v# Run with coverage informationpytest --cov=pydantic_prompt
You can run all formatting and linting checks with:
# Run all linting and formatting./lint.shOr run individual commands:
# Format coderuff format.# Check coderuff check.# Auto-fix issuesruff check --fix.# Type checkmypy pydantic_prompt
MIT
About
A simple library to document Pydantic models for structured LLM outputs using standard Python docstrings.