Skip to main content
This is an interactive notebook. You can run it locally or use the links below:

​
Leaderboard Quickstart

This quickstart shows you how to use the W&B Weave Leaderboard to compare model performance across multiple datasets and scoring functions. By the end, you will have published a leaderboard that ranks several models against a shared set of evaluations. You can then identify which model performs best on each metric. This guide is for developers familiar with running Weave evaluations who want to compare results side by side. Specifically, you will:
  1. Generate a dataset of fake zip code data.
  2. Author some scoring functions and evaluate a baseline model.
  3. Use these techniques to evaluate a matrix of models versus evaluations.
  4. Review the leaderboard in the Weave UI.

​
Step 1: Generate a dataset of fake zip code data

First, create a function generate_dataset_rows that generates a list of fake zip code data. This synthetic dataset gives the leaderboard a consistent set of inputs and expected values to score each model against. 
import json

from openai import OpenAI
from pydantic import BaseModel

class Row(BaseModel):
    zip_code: str
    city: str
    state: str
    avg_temp_f: float
    population: int
    median_income: int
    known_for: str

class Rows(BaseModel):
    rows: list[Row]

def generate_dataset_rows(
    location: str = "United States", count: int = 5, year: int = 2022
):
    client = OpenAI()

    completion = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[
            {"role": "system", "content": "You are a helpful assistant."},
            {
                "role": "user",
                "content": f"Please generate {count} rows of data for random zip codes in {location} for the year {year}.",
            },
        ],
        response_format={
            "type": "json_schema",
            "json_schema": {
                "name": "response_format",
                "schema": Rows.model_json_schema(),
            },
        },
    )

    return json.loads(completion.choices[0].message.content)["rows"]
python
import weave

weave.init("leaderboard-demo")

​
Step 2: Author scoring functions

Next, author three scoring functions. Each scorer evaluates a different aspect of the model output so the leaderboard can rank models on distinct dimensions of quality:
  1. check_concrete_fields: Checks if the model output matches the expected city and state.
  2. check_value_fields: Checks if the model output is within 10% of the expected population and median income.
  3. check_subjective_fields: Uses an LLM to check if the model output matches the expected “known for” field.
@weave.op
def check_concrete_fields(city: str, state: str, output: dict):
    return {
        "city_match": city == output["city"],
        "state_match": state == output["state"],
    }

@weave.op
def check_value_fields(
    avg_temp_f: float, population: int, median_income: int, output: dict
):
    return {
        "avg_temp_f_err": abs(avg_temp_f - output["avg_temp_f"]) / avg_temp_f,
        "population_err": abs(population - output["population"]) / population,
        "median_income_err": abs(median_income - output["median_income"])
        / median_income,
    }

@weave.op
def check_subjective_fields(zip_code: str, known_for: str, output: dict):
    client = OpenAI()

    class Response(BaseModel):
        correct_known_for: bool

    completion = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[
            {"role": "system", "content": "You are a helpful assistant."},
            {
                "role": "user",
                "content": f"My student was asked what the zip code {zip_code} is best known best for. The right answer is '{known_for}', and they said '{output['known_for']}'. Is their answer correct?",
            },
        ],
        response_format={
            "type": "json_schema",
            "json_schema": {
                "name": "response_format",
                "schema": Response.model_json_schema(),
            },
        },
    )

    return json.loads(completion.choices[0].message.content)

​
Step 3: Create an evaluation

Next, define an evaluation using the fake data and scoring functions. The Evaluation object pairs the dataset with the scorers, so you can run any model against the same benchmark. 
rows = generate_dataset_rows()
evaluation = weave.Evaluation(
    name="United States - 2022",
    dataset=rows,
    scorers=[
        check_concrete_fields,
        check_value_fields,
        check_subjective_fields,
    ],
)

​
Step 4: Evaluate a baseline model

Now, evaluate a baseline model that returns a static response. Establishing a baseline gives you a reference point on the leaderboard so you can measure how much each subsequent model improves over a static implementation. 
@weave.op
def baseline_model(zip_code: str):
    return {
        "city": "New York",
        "state": "NY",
        "avg_temp_f": 50.0,
        "population": 1000000,
        "median_income": 100000,
        "known_for": "The Big Apple",
    }

await evaluation.evaluate(baseline_model)

​
Step 5: Create more models

Now, create two more models to compare against the baseline. One model receives the zip code with no additional prompting, while the other receives a structured prompt. Comparing them on the leaderboard shows the impact of prompt context on answer quality. 
@weave.op
def gpt_4o_mini_no_context(zip_code: str):
    client = OpenAI()

    completion = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[{"role": "user", "content": f"""Zip code {zip_code}"""}],
        response_format={
            "type": "json_schema",
            "json_schema": {
                "name": "response_format",
                "schema": Row.model_json_schema(),
            },
        },
    )

    return json.loads(completion.choices[0].message.content)

await evaluation.evaluate(gpt_4o_mini_no_context)
python
@weave.op
def gpt_4o_mini_with_context(zip_code: str):
    client = OpenAI()

    completion = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[
            {
                "role": "user",
                "content": f"""Please answer the following questions about the zip code {zip_code}:
                   1. What is the city?
                   2. What is the state?
                   3. What is the average temperature in Fahrenheit?
                   4. What is the population?
                   5. What is the median income?
                   6. What is the most well known thing about this zip code?
                   """,
            }
        ],
        response_format={
            "type": "json_schema",
            "json_schema": {
                "name": "response_format",
                "schema": Row.model_json_schema(),
            },
        },
    )

    return json.loads(completion.choices[0].message.content)

await evaluation.evaluate(gpt_4o_mini_with_context)

​
Step 6: Create more evaluations

Now, evaluate a matrix of models versus evaluations. Running every model against several datasets (different regions and years) produces the data the leaderboard needs to rank models across multiple conditions. 
scorers = [
    check_concrete_fields,
    check_value_fields,
    check_subjective_fields,
]
evaluations = [
    weave.Evaluation(
        name="United States - 2022",
        dataset=weave.Dataset(
            name="United States - 2022",
            rows=generate_dataset_rows("United States", 5, 2022),
        ),
        scorers=scorers,
    ),
    weave.Evaluation(
        name="California - 2022",
        dataset=weave.Dataset(
            name="California - 2022", rows=generate_dataset_rows("California", 5, 2022)
        ),
        scorers=scorers,
    ),
    weave.Evaluation(
        name="United States - 2000",
        dataset=weave.Dataset(
            name="United States - 2000",
            rows=generate_dataset_rows("United States", 5, 2000),
        ),
        scorers=scorers,
    ),
]
models = [
    baseline_model,
    gpt_4o_mini_no_context,
    gpt_4o_mini_with_context,
]

for evaluation in evaluations:
    for model in models:
        await evaluation.evaluate(
            model, __weave={"display_name": evaluation.name + ":" + model.__name__}
        )

​
Step 7: Review the leaderboard

With the evaluation results published, you can now assemble them into a leaderboard for side-by-side comparison. You can create a new leaderboard by navigating to the leaderboard tab in the UI and clicking Create Leaderboard. You can also generate a leaderboard directly from Python: 
from weave.flow import leaderboard
from weave.trace.ref_util import get_ref

spec = leaderboard.Leaderboard(
    name="Zip Code World Knowledge",
    description="""
This leaderboard compares the performance of models in terms of world knowledge about zip codes.

### Columns

1. **State Match against `United States - 2022`**: The fraction of zip codes that the model correctly identified the state for.
2. **Avg Temp F Error against `California - 2022`**: The mean absolute error of the model's average temperature prediction.
3. **Correct Known For against `United States - 2000`**: The fraction of zip codes that the model correctly identified the most well known thing about the zip code.
""",
    columns=[
        leaderboard.LeaderboardColumn(
            evaluation_object_ref=get_ref(evaluations[0]).uri(),
            scorer_name="check_concrete_fields",
            summary_metric_path="state_match.true_fraction",
        ),
        leaderboard.LeaderboardColumn(
            evaluation_object_ref=get_ref(evaluations[1]).uri(),
            scorer_name="check_value_fields",
            should_minimize=True,
            summary_metric_path="avg_temp_f_err.mean",
        ),
        leaderboard.LeaderboardColumn(
            evaluation_object_ref=get_ref(evaluations[2]).uri(),
            scorer_name="check_subjective_fields",
            summary_metric_path="correct_known_for.true_fraction",
        ),
    ],
)

ref = weave.publish(spec)
You now have a published leaderboard in Weave that ranks each model against the three evaluations and scoring metrics you defined. From the Weave UI, you can inspect per-model scores, drill into individual evaluation runs, and compare future model iterations against the same baseline.