Bring Your Own Metric

NeMo Platform offers built-in metrics that can be configured to evaluate on your custom data. You can bring your own metric into the NeMo Platform ecosystem with remote metrics.

A remote metric can integrate with your custom evaluation served by a REST API. You have full control of the logic and evaluation that executes and the reported scores.

Overview

Remote metrics support two types:

Type	Use Case	Payload Structure
Generic Remote (remote)	Custom endpoints with configurable body/scores	User-defined Jinja template
NeMo Agent Toolkit Remote (nemo-agent-toolkit-remote)	NAT evaluator endpoints	Fixed: {evaluator_name, item}

NeMo Evaluator supports two evaluation modes:

Mode	Use Case	Dataset Size	Response
Live Evaluation	Rapid prototyping, testing	Up to 10 rows	Immediate (synchronous)
Job Evaluation	Production workloads, full datasets	Unlimited	Async (poll for completion)

Prerequisites

Before running remote metric evaluations:

1. Workspace: Have a workspace created.

2. Remote endpoint: Have your evaluation endpoint running and accessible.

3. API key (if required): If your endpoint requires authentication, create a secret to store the API key.

4. Initialize the SDK:

import os
from nemo_platform import NeMoPlatform
from nemo_platform.types.evaluation import (
    EvaluateDatasetRowsParam,
    RemoteMetricParam,
    NeMoAgentToolkitRemoteMetricParam,
)

client = NeMoPlatform(
    base_url=os.environ.get("NMP_BASE_URL", "http://localhost:8080"),
    workspace="default",
)

---

Live Evaluation

Live evaluation provides immediate results for rapid iteration when developing and testing your metrics.

Generic Remote Metric

Use a generic remote metric when you need full control over the request payload and score extraction:

metric: RemoteMetricParam = {
    "type": "remote",
    "url": "https://my-evaluation-server.test/evaluate",
    "body": {
        "reference": "{{ item.reference }}",
        "response": "{{ item.output }}"
    },
    "scores": [
        {
            "name": "accuracy",
            "parser": {"type": "json", "json_path": "$.result.accuracy"}
        }
    ],
    "timeout_seconds": 30.0,
    "max_retries": 3,
}

dataset: EvaluateDatasetRowsParam = {
    "rows": [
        {"reference": "The capital is Paris", "output": "Paris is the capital"},
        {"reference": "2", "output": "2"},
    ]
}

result = client.evaluation.metrics.evaluate(
    metric=metric,
    dataset=dataset,
)

# Access results
for score in result.aggregate_scores:
    print(f"{score.name}: mean={score.mean}, count={score.count}")

Key configuration:

• body: Jinja template for the request payload. Use {{ item.<column> }} to access dataset columns.

• scores: List of score definitions with a parser object containing JSONPath expression for extracting values from the response.

NeMo Agent Toolkit Remote Metric

Use the NAT remote metric type when integrating with NeMo Agent Toolkit evaluators:

metric: NeMoAgentToolkitRemoteMetricParam = {
    "type": "nemo-agent-toolkit-remote",
    "url": "http://localhost:8001/evaluate_item",
    "evaluator_name": "similarity_eval",
    "timeout_seconds": 30.0,
    "max_retries": 3,
}

dataset: EvaluateDatasetRowsParam = {
    "rows": [
        {
            "id": "item_1",
            "input_obj": "What is the capital of France?",
            "expected_output_obj": "The capital of France is Paris.",
            "output_obj": "Paris is the capital of France.",
            "trajectory": [],
            "expected_trajectory": [],
            "full_dataset_entry": {},
        }
    ]
}

result = client.evaluation.metrics.evaluate(
    metric=metric,
    dataset=dataset,
)

print(f"Score: {result.aggregate_scores[0].mean}")

The NAT metric automatically:

• Sends payload: {"evaluator_name": "<name>", "item": <row_data>}

• Extracts score from: $.result.score

---

Job-Based Evaluation

For larger datasets or production workloads, use job-based evaluation. Jobs run asynchronously and support datasets of any size.

Create a Job with Inline Metric

Generic Remote Metric

from nemo_platform.types.evaluation import (
    RemoteMetricParam,
    MetricOfflineJobParam,
)

metric: RemoteMetricParam = {
    "type": "remote",
    "url": "https://my-evaluation-server.test/evaluate",
    "body": {
        "reference": "{{ item.reference }}",
        "response": "{{ item.output }}"
    },
    "scores": [
        {
            "name": "accuracy",
            "parser": {"type": "json", "json_path": "$.result.accuracy"}
        }
    ],
    "timeout_seconds": 30.0,
    "max_retries": 3,
}

job = client.evaluation.metric_jobs.create(
    spec=MetricOfflineJobParam(
        metric=metric,
        dataset={
            "rows": [
                {"reference": "Paris", "output": "Paris"},
                {"reference": "2", "output": "2"},
            ]
        },
    ),
)

print(f"Job created: {job.name} ({job.id})")

NAT Remote Metric

from nemo_platform.types.evaluation import (
    NeMoAgentToolkitRemoteMetricParam,
    MetricOfflineJobParam,
)

metric: NeMoAgentToolkitRemoteMetricParam = {
    "type": "nemo-agent-toolkit-remote",
    "url": "http://host.docker.internal:8001/evaluate_item",
    "evaluator_name": "similarity_eval",
    "timeout_seconds": 30.0,
    "max_retries": 3,
}

job = client.evaluation.metric_jobs.create(
    spec=MetricOfflineJobParam(
        metric=metric,
        dataset={
            "rows": [
                {
                    "id": "item_1",
                    "input_obj": "What is the capital of France?",
                    "expected_output_obj": "The capital of France is Paris.",
                    "output_obj": "Paris is the capital.",
                    "trajectory": [],
                    "expected_trajectory": [],
                    "full_dataset_entry": {},
                }
            ]
        },
    ),
)

print(f"Job created: {job.name} ({job.id})")

Create a Stored Metric

You can create a reusable metric and reference it by name in jobs:

# Create the metric
client.evaluation.metrics.create(
    type="remote",
    name="my-remote-metric",
    description="Custom evaluation metric for accuracy scoring",
    url="https://my-evaluation-server.test/evaluate",
    body={"reference": "{{ item.reference }}", "response": "{{ item.output }}"},
    scores=[{"name": "accuracy", "parser": {"type": "json", "json_path": "$.result.accuracy"}}],
)

# Use it in a job by reference (workspace/metric-name)
job = client.evaluation.metric_jobs.create(
    spec={"metric": "default/my-remote-metric", "dataset": "default/my-dataset-fileset"},
)

Refer to Managing Metrics for more information on how to modify or delete a metric.

Monitor Job Progress

import time

while True:
    job_status = client.evaluation.metric_jobs.retrieve(job.name)
    print(f"Status: {job_status.status}")

    if job_status.status in ["completed", "error", "cancelled"]:
        break

    time.sleep(5)

---

Using API Key Secrets

If your remote endpoint requires authentication, store the API key as a secret:

Create a Secret

client.secrets.create(
    name="my-remote-api-key",
    data="your-api-key-value"
)

Reference the Secret in Your Metric

from nemo_platform.types.evaluation import (
    RemoteMetricParam,
    NeMoAgentToolkitRemoteMetricParam,
)

# Live evaluation with secret
metric: RemoteMetricParam = {
    "type": "remote",
    "url": "https://my-authenticated-endpoint.test/evaluate",
    "body": {"input": "{{ item.input }}"},
    "scores": [{"name": "score", "parser": {"type": "json", "json_path": "$.score"}}],
    "api_key_secret": "my-remote-api-key",
}

result = client.evaluation.metrics.evaluate(
    metric=metric,
    dataset={"rows": [{"input": "test"}]},
)

# Job evaluation with secret
nat_metric: NeMoAgentToolkitRemoteMetricParam = {
    "type": "nemo-agent-toolkit-remote",
    "url": "http://host.docker.internal:8001/evaluate_item",
    "evaluator_name": "similarity_eval",
    "api_key_secret": "my-remote-api-key",
}

job = client.evaluation.metric_jobs.create(
    spec={"metric": nat_metric, "dataset": {"rows": [...]}},
)

The secret is automatically resolved:

• Live evaluation: Secret is fetched from the platform’s secrets service

• Job evaluation: Secret is injected as an environment variable into the container

The API key is sent in the Authorization: Bearer <key> header.

---

Endpoint Requirements

Your remote endpoint must:

1. Accept POST requests with Content-Type: application/json

2. Return a JSON response containing the score(s)

Example Endpoint (FastAPI)

from fastapi import FastAPI
from pydantic import BaseModel

app = FastAPI()

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

class EvaluationResponse(BaseModel):
    result: dict

@app.post("/evaluate")
async def evaluate(request: EvaluationRequest) -> EvaluationResponse:
    # Your evaluation logic here
    accuracy = 1.0 if request.reference == request.response else 0.0
    return EvaluationResponse(result={"accuracy": accuracy})

NAT Endpoint Format

NAT endpoints receive:

{
    "evaluator_name": "similarity_eval",
    "item": {
        "id": "item_1",
        "input_obj": "What is the capital of France?",
        "expected_output_obj": "The capital of France is Paris.",
        "output_obj": "Paris is the capital.",
        "trajectory": [],
        "expected_trajectory": [],
        "full_dataset_entry": {}
    }
}

And must return:

{
    "success": true,
    "result": {
        "id": "item_1",
        "score": 0.85,
        "reasoning": {"method": "cosine_similarity"}
    },
    "error": null
}

---

Configuration Options

Metric Parameters

Parameter	Type	Description
type	string	"remote" or "nemo-agent-toolkit-remote"
url	string	Endpoint URL
body	dict	(Generic only) Jinja template for request payload
scores	list	(Generic only) List of score configuration objects (refer to Score Configuration section)
evaluator_name	string	(NAT only) Name of the NAT evaluator
api_key_secret	string	Optional secret name for API key authentication
timeout_seconds	float	Request timeout (default: 30.0)
max_retries	int	Max retry attempts (default: 3)

Score Configuration (Generic Remote Only)

Each score object in the scores list supports the following fields:

Field	Type	Required	Description
name	string	Yes	Score identifier (lowercase, numbers, underscores only)
parser	object	Yes	Parser configuration for extracting the score value (refer to Parser configuration section)
description	string	No	Human-readable description of the score
minimum	float	No	Minimum expected value for the score range (default: None = no bound)
maximum	float	No	Maximum expected value for the score range (default: None = no bound)

Parser configuration:

Field	Type	Required	Description
type	string	Yes	Parser type, must be "json"
json_path	string	Yes	JSONPath expression to extract the score value

Example with all fields:

{
    "scores": [
        {
            "name": "accuracy",
            "parser": {
                "type": "json",
                "json_path": "$.result.accuracy"
            },
            "description": "Measures response accuracy against reference",
            "minimum": 0.0,
            "maximum": 1.0
        }
    ]
}

---

Job Management

After successfully creating a job, refer to Metrics Job Management to oversee execution and monitor progress.

---

Limitations

1. Network access: For job-based evaluation, endpoints must be accessible from the job container. Use host.docker.internal for local endpoints.

2. Response format: Scores must be extractable using JSONPath from the response. Ensure your endpoint returns properly structured JSON.

3. Live evaluation limits: Live evaluations are limited to 10 rows. Use job-based evaluation for larger datasets.

See also

• Evaluation Results - Understanding and downloading results

• LLM-as-a-Judge - Use an LLM to evaluate outputs

• Agentic Evaluation - Evaluate agent workflows