EvalTask
One declarative eval task.
Example Usage
typescript
import { EvalTask } from "@meetkai/mka1/models/components";
let value: EvalTask = {
id: "<id>",
type: "math",
dataset: {},
promptTemplate: "<value>",
grader: {
type: "python",
},
};Fields
| Field | Type | Required | Description |
|---|---|---|---|
id | string | ✔️ | N/A |
name | string | ➖ | N/A |
type | components.EvalTaskType | ✔️ | N/A |
dataset | components.EvalDataset | ✔️ | Dataset backing an eval task. |
promptTemplate | string | ✔️ | N/A |
targetTemplate | string | ➖ | N/A |
choices | string[] | ➖ | N/A |
outputExtraction | components.EvalOutputExtraction | ➖ | N/A |
metrics | any[] | ➖ | N/A |
grader | components.EvalPythonGrader | ✔️ | N/A |
preprocess | components.EvalPythonPreprocessor | ➖ | N/A |
fewshot | components.EvalFewshotConfig | ➖ | N/A |
numFewshot | number | ➖ | lm-eval style alias for fewshot.count. |
metadata | Record<string, any> | ➖ | N/A |