Implementation:Explodinggradients Ragas MetricBasePrompt Class

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

Overview

The BasePrompt class in ragas.prompt.metrics.base_prompt is a generic abstract base for structured metric prompts with type-safe Pydantic input/output models and built-in internationalization (i18n) support through language adaptation.

Description

BasePrompt[InputModel, OutputModel] is an abstract generic class that defines class-level attributes: input_model, output_model, instruction, examples (list of input/output tuples), and language (default "english"). The to_string method assembles a complete prompt by combining the instruction, a JSON Schema specification for the output model, formatted examples, and the serialized input data. The _generate_examples helper formats example pairs into numbered text blocks. The adapt async method translates all strings in the examples to a target language using a private _translate_strings helper that calls an InstructorBaseRagasLLM with a structured _TranslatedStrings response model and safety-focused translation instructions. Optionally, the instruction itself can also be translated. The method returns a deep copy of the prompt with updated examples and language.

Usage

Subclass BasePrompt with concrete Pydantic input/output models and an instruction. Use to_string(data) to render the complete prompt for an LLM. Call adapt(target_language, llm) to produce a translated variant.

Code Reference

Item	Detail
Source Location	`src/ragas/prompt/metrics/base_prompt.py` L78-193
Class Signature	`class BasePrompt(ABC, Generic[InputModel, OutputModel])`
Key Methods	`to_string(data)`, `_generate_examples()`, `adapt(target_language, llm, adapt_instruction)`
Import	`from ragas.prompt.metrics.base_prompt import BasePrompt`

I/O Contract

Inputs

Parameter	Type	Description
`input_model` (class attr)	`Type[InputModel]`	Pydantic model class for input validation
`output_model` (class attr)	`Type[OutputModel]`	Pydantic model class for output schema generation
`instruction` (class attr)	`str`	Task description for the LLM
`examples` (class attr)	`List[Tuple[InputModel, OutputModel]]`	Few-shot example pairs
`data` (to_string)	`InputModel`	Input data instance to render into the prompt
`target_language` (adapt)	`str`	Target language for translation (e.g., "spanish", "hindi")

Outputs

Method	Return Type	Description
`to_string()`	`str`	Complete prompt string ready for LLM submission
`adapt()`	`BasePrompt`	Deep copy of the prompt with translated examples and language

Usage Examples

from pydantic import BaseModel
from ragas.prompt.metrics.base_prompt import BasePrompt

class EvalInput(BaseModel):
    response: str
    reference: str

class EvalOutput(BaseModel):
    score: float
    explanation: str

class CorrectnessPrompt(BasePrompt[EvalInput, EvalOutput]):
    input_model = EvalInput
    output_model = EvalOutput
    instruction = "Evaluate the correctness of the response against the reference."
    examples = [
        (
            EvalInput(response="Paris is the capital of France.", reference="The capital of France is Paris."),
            EvalOutput(score=1.0, explanation="The response correctly identifies Paris as the capital."),
        )
    ]

prompt = CorrectnessPrompt()
prompt_text = prompt.to_string(
    EvalInput(response="Berlin is the capital of Germany.", reference="Germany's capital is Berlin.")
)

# Adapt to Spanish
spanish_prompt = await prompt.adapt("spanish", llm=instructor_llm)

Related Pages

PydanticPrompt_Class - Similar structured prompting in the main prompt module
PromptUtils_Module - get_all_strings/update_strings used by adapt()
BasePrompt_Class - The base prompt in ragas.prompt.base (different class)

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment