​
Data Models Reference

This document describes the core data models used in the Reward Kit for representing messages, evaluation results, and metrics.

​
Message Models

​
Message

The Message class represents a single message in a conversation.

from reward_kit import Message

message = Message(
    role="assistant",
    content="This is the response content",
    name=None,  # Optional
    tool_call_id=None,  # Optional
    tool_calls=None,  # Optional
    function_call=None  # Optional
)

​
Attributes

  • role (str): The role of the message sender. Typically one of:

    • "user": Message from the user
    • "assistant": Message from the assistant
    • "system": System message providing context/instructions

  • content (str): The text content of the message.

  • name (Optional[str]): Optional name of the sender (for named system messages).

  • tool_call_id (Optional[str]): Optional ID for a tool call (used in tool calling).

  • tool_calls (Optional[List[Dict[str, Any]]]): Optional list of tool calls in the message.

  • function_call (Optional[Dict[str, Any]]): Optional function call information (legacy format).

​
Compatibility

The Message class is compatible with OpenAI’s ChatCompletionMessageParam interface, allowing for easy integration with OpenAI-compatible APIs.

​
Evaluation Models

​
EvaluateResult

The EvaluateResult class represents the complete result of an evaluator with multiple metrics.

from reward_kit import EvaluateResult, MetricResult

result = EvaluateResult(
    score=0.75,
    reason="Overall good response with minor issues",
    metrics={
        "clarity": MetricResult(score=0.8, reason="Clear and concise", success=True),
        "accuracy": MetricResult(score=0.7, reason="Contains a minor factual error", success=True)
    },
    error=None  # Optional error message
)

​
Attributes

  • score (float): The overall evaluation score, typically between 0.0 and 1.0.

  • reason (Optional[str]): Optional explanation for the overall score.

  • metrics (Dict[str, MetricResult]): Dictionary of component metrics.

  • error (Optional[str]): Optional error message if the evaluation encountered a problem.

​
MetricResult

The MetricResult class represents a single metric in an evaluation.

from reward_kit import MetricResult

metric = MetricResult(
    score=0.8,
    reason="The response provides a clear explanation with appropriate examples",
    success=True
)

​
Attributes

  • score (float): The score for this specific metric, typically between 0.0 and 1.0.

  • reason (str): Explanation for why this score was assigned.

  • success (bool): Indicates whether the metric condition was met (e.g., pass/fail).

​
Example Usages

​
Working with Messages

from reward_kit import Message

# Create a user message
user_message = Message(
    role="user",
    content="Can you explain how machine learning works?"
)

# Create an assistant message
assistant_message = Message(
    role="assistant",
    content="Machine learning is a method where computers learn from data without being explicitly programmed."
)

# Create a system message
system_message = Message(
    role="system",
    content="You are a helpful assistant that provides clear and accurate explanations."
)

# Create a message with tool calls
tool_call_message = Message(
    role="assistant",
    content=None,
    tool_calls=[{
        "id": "call_123",
        "type": "function",
        "function": {
            "name": "get_weather",
            "arguments": '{"location": "San Francisco", "unit": "celsius"}'
        }
    }]
)

​
Working with EvaluateResult

from reward_kit import EvaluateResult, MetricResult

# Create an EvaluateResult
eval_result = EvaluateResult(
    score=0.75,
    reason="Overall good response with some minor issues",
    metrics={
        "clarity": MetricResult(score=0.8, reason="Clear and concise explanation", success=True),
        "accuracy": MetricResult(score=0.7, reason="Contains one minor factual error", success=True),
        "relevance": MetricResult(score=0.75, reason="Mostly relevant to the query", success=True)
    }
)

# Access metrics
clarity_score = eval_result.metrics["clarity"].score
print(f"Clarity score: {clarity_score}")  # Clarity score: 0.8

# Check for errors
if eval_result.error:
    print(f"Evaluation error: {eval_result.error}")
else:
    print(f"Evaluation successful with score: {eval_result.score}")

​
Type Compatibility

While the classes provide strong typing for development, the Reward Kit also accepts dictionary representations for flexibility:

# Using dictionaries instead of Message objects
messages = [
    {"role": "user", "content": "What is machine learning?"},
    {"role": "assistant", "content": "Machine learning is a method..."}
]

# These are automatically converted to the appropriate types internally

This flexibility makes it easier to integrate with different APIs and data formats.