Implementation:Explodinggradients Ragas PydanticPrompt Class

Field	Value
source	Explodinggradients_Ragas (GitHub)
domains	Prompts, Framework
last_updated	2026-02-10 00:00 GMT

Overview

The PydanticPrompt class provides structured, type-safe prompting with Pydantic input/output models, supporting generation, output parsing with retries, language adaptation, and JSON serialization.

Description

PydanticPrompt[InputModel, OutputModel] is a generic class that extends BasePrompt. It defines class-level attributes for input_model, output_model, instruction, and examples. The to_string method assembles a complete prompt string including the instruction, JSON schema for the output model, formatted examples, and serialized input data. The generate and generate_multiple methods handle three LLM backends: standard BaseRagasLLM, InstructorBaseRagasLLM, and legacy LangChain LLMs. Output parsing is handled by RagasOutputParser with configurable retry logic via FixOutputFormat. The adapt method translates prompt examples to a target language using the TranslateStatements prompt. The class provides save/load methods for JSON persistence with hash-based integrity verification.

Usage

Subclass PydanticPrompt with concrete Pydantic models for input and output, set the instruction and optionally examples, then call generate() with an LLM and input data. Override process_input and process_output for custom pre/post-processing.

Code Reference

Item	Detail
Source Location	src/ragas/prompt/pydantic_prompt.py L82-627
Class Signature	class PydanticPrompt(BasePrompt, Generic[InputModel, OutputModel])
Key Methods	generate(), generate_multiple(), adapt(), to_string(), save(), load()
Import	from ragas.prompt import PydanticPrompt

I/O Contract

Inputs (generate)

Parameter	Type	Description
llm	Union[BaseRagasLLM, InstructorBaseRagasLLM, BaseLanguageModel]	Language model instance
data	InputModel	Pydantic model instance with input data
temperature	Optional[float]	Generation temperature
stop	Optional[List[str]]	Stop sequences
callbacks	Optional[Callbacks]	Callback handlers
retries_left	int	Number of retry attempts for invalid output (default 3)

Outputs

Method	Return Type	Description
generate()	OutputModel	Single parsed output model instance
generate_multiple()	List[OutputModel]	List of n parsed output model instances
adapt()	PydanticPrompt	New prompt with translated examples
to_string()	str	Complete prompt string with schema and examples

Usage Examples

from pydantic import BaseModel
from ragas.prompt import PydanticPrompt

class QuestionInput(BaseModel):
    context: str
    question: str

class AnswerOutput(BaseModel):
    answer: str
    confidence: float

class QAPrompt(PydanticPrompt[QuestionInput, AnswerOutput]):
    instruction = "Answer the question based on the context."
    input_model = QuestionInput
    output_model = AnswerOutput
    examples = [
        (
            QuestionInput(context="The sky is blue.", question="What color is the sky?"),
            AnswerOutput(answer="Blue", confidence=0.95),
        )
    ]

prompt = QAPrompt()
result = await prompt.generate(
    llm=my_llm,
    data=QuestionInput(context="Water boils at 100C.", question="At what temperature does water boil?"),
)
print(result.answer)       # "100 degrees Celsius"
print(result.confidence)   # 0.92

Related Pages

• BasePrompt_Class - Parent abstract class
• FewShotPydanticPrompt_Class - Dynamic few-shot variant
• ImageTextPrompt_Class - Multi-modal extension
• PromptUtils_Module - String extraction utilities used by adapt()