Datasets:
roanbrasil
/
devops-gitops-corpus

Dataset card Data Studio Files
xet
 Community
1
text
stringlengths
0
59.1k
### Combine Multiple Scorers
Use complementary scorers to catch different failure modes:
```ts
scorers: [
scorers.exactMatch, // strict match
scorers.embeddingSimilarity, // semantic similarity
scorers.moderation, // safety check
];
```
### Set Realistic Thresholds
Start with loose thresholds and tighten over time:
```ts
scorers: [
{ scorer: scorers.answerCorrectness, threshold: 0.7 }, // initial baseline
];
```
Monitor false positives and adjust based on production data.
### Tag Runs for Filtering
Use tags to organize runs by context:
```ts
tags: [`branch:${process.env.GIT_BRANCH}`, `commit:${process.env.GIT_SHA}`, "ci"];
```
This enables filtering in VoltOps dashboards and APIs.
### Handle Long-Running Items
Set timeouts in your runner to prevent hangs:
```ts
runner: async ({ item, signal }) => {
const timeout = setTimeout(() => controller.abort(), 30000);
try {
return await agent.generateText(item.input, { signal });
} finally {
clearTimeout(timeout);
}
};
```
Or use the provided `signal` for coordinated cancellation.
### Validate Experiment Configuration
Test your experiment with a small dataset before running the full suite:
```ts
const result = await runExperiment(experiment, {
voltOpsClient,
onProgress: ({ completed, total }) => {
if (completed === 1) {
console.log("First item processed successfully");
}
},
});
```
Check the first result to verify runner output format and scorer compatibility.
## Examples
### Basic Regression Test
```ts
import { createExperiment } from "@voltagent/evals";
import { scorers } from "@voltagent/scorers";
export default createExperiment({
id: "greeting-smoke",
dataset: {
items: [
{ id: "1", input: "hello", expected: "hello" },
{ id: "2", input: "goodbye", expected: "goodbye" },
],
},
runner: async ({ item }) => item.input.toLowerCase(),
scorers: [scorers.exactMatch],
passCriteria: { type: "passRate", min: 1.0 },
});
```
### RAG Evaluation
```ts
import { createExperiment } from "@voltagent/evals";
import { scorers } from "@voltagent/scorers";
export default createExperiment({
id: "rag-quality",
dataset: { name: "knowledge-base-qa" },