Implementation:Explodinggradients Ragas FewShotPydanticPrompt Class

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

Overview

The FewShotPydanticPrompt class extends PydanticPrompt to dynamically select the most relevant Pydantic-typed examples at generation time using an embeddings-based ExampleStore.

Description

FewShotPydanticPrompt[InputModel, OutputModel] is a dataclass that wraps a PydanticPrompt with an ExampleStore for dynamic example retrieval. The ExampleStore abstract class defines get_examples and add_example interfaces operating on Pydantic BaseModel instances. InMemoryExampleStore implements this using embeddings: it serializes input models to JSON for embedding generation, stores embeddings in a list, and retrieves the top-k most similar examples above a configurable similarity threshold using NumPy-based cosine similarity. The generate_multiple override retrieves relevant examples from the store before delegating to the parent PydanticPrompt.generate_multiple. The from_pydantic_prompt class method converts an existing PydanticPrompt into a few-shot variant by transferring all examples into a new InMemoryExampleStore.

Usage

Use FewShotPydanticPrompt.from_pydantic_prompt() to convert an existing prompt, or instantiate directly with an ExampleStore. Call add_example() to grow the example pool. Each generate() call dynamically selects the best examples for the given input.

Code Reference

Item	Detail
Source Location	src/ragas/prompt/few_shot_pydantic_prompt.py L97-155
Classes	ExampleStore(ABC), InMemoryExampleStore, FewShotPydanticPrompt(PydanticPrompt, Generic[InputModel, OutputModel])
Key Methods	generate_multiple(), add_example(), from_pydantic_prompt()
Import	from ragas.prompt.few_shot_pydantic_prompt import FewShotPydanticPrompt

I/O Contract

Inputs

Parameter	Type	Description
example_store	ExampleStore	Store for managing and retrieving examples
top_k_for_examples	int	Number of examples to retrieve (default 5)
threshold_for_examples	float	Minimum similarity threshold (default 0.7)
llm (generate)	BaseRagasLLM	Language model for generation
data (generate)	InputModel	Pydantic input data for example retrieval and generation

Outputs

Method	Return Type	Description
generate()	OutputModel	Single parsed output with dynamically chosen examples
generate_multiple()	List[OutputModel]	Multiple parsed outputs
from_pydantic_prompt()	FewShotPydanticPrompt	Converted prompt with embedded example store

Usage Examples

from ragas.prompt.few_shot_pydantic_prompt import FewShotPydanticPrompt, InMemoryExampleStore
from ragas.prompt import PydanticPrompt
from pydantic import BaseModel

class QInput(BaseModel):
    question: str

class QOutput(BaseModel):
    answer: str

class MyPrompt(PydanticPrompt[QInput, QOutput]):
    instruction = "Answer the question."
    input_model = QInput
    output_model = QOutput
    examples = [
        (QInput(question="What is 2+2?"), QOutput(answer="4")),
        (QInput(question="What is H2O?"), QOutput(answer="Water")),
    ]

# Convert existing prompt to few-shot
few_shot = FewShotPydanticPrompt.from_pydantic_prompt(
    MyPrompt(), embeddings=my_embedding_model
)

# Add more examples dynamically
few_shot.add_example(QInput(question="What is pi?"), QOutput(answer="3.14159"))

# Generate with dynamic example selection
result = await few_shot.generate(llm=my_llm, data=QInput(question="What is e?"))
print(result.answer)

Related Pages

• PydanticPrompt_Class - Parent class providing core generation logic
• DynamicFewShotPrompt_Class - Dict-based dynamic few-shot variant
• BasePrompt_Class - Root of the prompt hierarchy