@cogitator-ai/evals 0.8.0 → 0.10.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +255 -54
- package/dist/assertions/error-rate.d.ts +8 -0
- package/dist/assertions/error-rate.d.ts.map +1 -0
- package/dist/assertions/error-rate.js +27 -0
- package/dist/assertions/error-rate.js.map +1 -0
- package/dist/assertions/index.d.ts +3 -0
- package/dist/assertions/index.d.ts.map +1 -1
- package/dist/assertions/index.js +1 -0
- package/dist/assertions/index.js.map +1 -1
- package/dist/assertions/regression.d.ts.map +1 -1
- package/dist/assertions/regression.js +3 -1
- package/dist/assertions/regression.js.map +1 -1
- package/dist/assertions/threshold.d.ts.map +1 -1
- package/dist/assertions/threshold.js +4 -1
- package/dist/assertions/threshold.js.map +1 -1
- package/dist/datasets/csv-loader.d.ts +5 -0
- package/dist/datasets/csv-loader.d.ts.map +1 -1
- package/dist/datasets/csv-loader.js +89 -7
- package/dist/datasets/csv-loader.js.map +1 -1
- package/dist/datasets/dataset.d.ts +11 -2
- package/dist/datasets/dataset.d.ts.map +1 -1
- package/dist/datasets/dataset.js +42 -13
- package/dist/datasets/dataset.js.map +1 -1
- package/dist/datasets/issues.d.ts +4 -0
- package/dist/datasets/issues.d.ts.map +1 -0
- package/dist/datasets/issues.js +7 -0
- package/dist/datasets/issues.js.map +1 -0
- package/dist/datasets/jsonl-loader.d.ts.map +1 -1
- package/dist/datasets/jsonl-loader.js +6 -1
- package/dist/datasets/jsonl-loader.js.map +1 -1
- package/dist/eval-builder.d.ts +3 -0
- package/dist/eval-builder.d.ts.map +1 -1
- package/dist/eval-builder.js +7 -0
- package/dist/eval-builder.js.map +1 -1
- package/dist/eval-comparison.d.ts +7 -0
- package/dist/eval-comparison.d.ts.map +1 -1
- package/dist/eval-comparison.js +42 -12
- package/dist/eval-comparison.js.map +1 -1
- package/dist/eval-suite.d.ts +55 -8
- package/dist/eval-suite.d.ts.map +1 -1
- package/dist/eval-suite.js +254 -53
- package/dist/eval-suite.js.map +1 -1
- package/dist/index.d.ts +14 -8
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +6 -2
- package/dist/index.js.map +1 -1
- package/dist/matchers.d.ts +68 -0
- package/dist/matchers.d.ts.map +1 -0
- package/dist/matchers.js +112 -0
- package/dist/matchers.js.map +1 -0
- package/dist/metrics/custom.d.ts +19 -7
- package/dist/metrics/custom.d.ts.map +1 -1
- package/dist/metrics/custom.js +5 -0
- package/dist/metrics/custom.js.map +1 -1
- package/dist/metrics/index.d.ts +8 -4
- package/dist/metrics/index.d.ts.map +1 -1
- package/dist/metrics/index.js +4 -2
- package/dist/metrics/index.js.map +1 -1
- package/dist/metrics/llm-judge.d.ts +20 -1
- package/dist/metrics/llm-judge.d.ts.map +1 -1
- package/dist/metrics/llm-judge.js +162 -20
- package/dist/metrics/llm-judge.js.map +1 -1
- package/dist/metrics/matching.d.ts +28 -0
- package/dist/metrics/matching.d.ts.map +1 -0
- package/dist/metrics/matching.js +124 -0
- package/dist/metrics/matching.js.map +1 -0
- package/dist/metrics/retrieval.d.ts +21 -0
- package/dist/metrics/retrieval.d.ts.map +1 -0
- package/dist/metrics/retrieval.js +108 -0
- package/dist/metrics/retrieval.js.map +1 -0
- package/dist/metrics/statistical.d.ts +28 -0
- package/dist/metrics/statistical.d.ts.map +1 -1
- package/dist/metrics/statistical.js +61 -0
- package/dist/metrics/statistical.js.map +1 -1
- package/dist/metrics/tool-calls.d.ts +43 -0
- package/dist/metrics/tool-calls.d.ts.map +1 -0
- package/dist/metrics/tool-calls.js +106 -0
- package/dist/metrics/tool-calls.js.map +1 -0
- package/dist/metrics/types.d.ts +53 -5
- package/dist/metrics/types.d.ts.map +1 -1
- package/dist/metrics/types.js.map +1 -1
- package/dist/reporters/ci.d.ts +4 -0
- package/dist/reporters/ci.d.ts.map +1 -1
- package/dist/reporters/ci.js +5 -1
- package/dist/reporters/ci.js.map +1 -1
- package/dist/reporters/console.d.ts.map +1 -1
- package/dist/reporters/console.js +11 -4
- package/dist/reporters/console.js.map +1 -1
- package/dist/reporters/csv.d.ts.map +1 -1
- package/dist/reporters/csv.js +5 -2
- package/dist/reporters/csv.js.map +1 -1
- package/dist/reporters/format.d.ts +19 -0
- package/dist/reporters/format.d.ts.map +1 -0
- package/dist/reporters/format.js +31 -0
- package/dist/reporters/format.js.map +1 -0
- package/dist/reporters/github.d.ts +11 -0
- package/dist/reporters/github.d.ts.map +1 -0
- package/dist/reporters/github.js +95 -0
- package/dist/reporters/github.js.map +1 -0
- package/dist/reporters/index.d.ts +44 -20
- package/dist/reporters/index.d.ts.map +1 -1
- package/dist/reporters/index.js +14 -3
- package/dist/reporters/index.js.map +1 -1
- package/dist/reporters/junit.d.ts +8 -0
- package/dist/reporters/junit.d.ts.map +1 -0
- package/dist/reporters/junit.js +107 -0
- package/dist/reporters/junit.js.map +1 -0
- package/dist/schema.d.ts +25 -0
- package/dist/schema.d.ts.map +1 -1
- package/dist/schema.js +29 -0
- package/dist/schema.js.map +1 -1
- package/dist/vitest.d.ts +35 -0
- package/dist/vitest.d.ts.map +1 -0
- package/dist/vitest.js +85 -0
- package/dist/vitest.js.map +1 -0
- package/package.json +17 -3
package/dist/vitest.js
ADDED
|
@@ -0,0 +1,85 @@
|
|
|
1
|
+
import { beforeAll, describe, it } from 'vitest';
|
|
2
|
+
import { caseLabel } from './reporters/format.js';
|
|
3
|
+
const DEFAULT_TIMEOUT = 10 * 60 * 1000;
|
|
4
|
+
/** Time the hook keeps after the deadline settled the outcome, for Vitest itself */
|
|
5
|
+
const HOOK_GRACE = 5 * 1000;
|
|
6
|
+
/** The longest delay `setTimeout` keeps, less the grace the hook adds to it */
|
|
7
|
+
const MAX_TIMEOUT = 2_147_483_647 - HOOK_GRACE;
|
|
8
|
+
function messageOf(error) {
|
|
9
|
+
return error instanceof Error ? error.message : String(error);
|
|
10
|
+
}
|
|
11
|
+
/**
|
|
12
|
+
* Registers a Vitest describe block for an eval suite. The suite runs once in `beforeAll`, each
|
|
13
|
+
* case becomes a test that fails when the case errored, and an `assertions` test fails with
|
|
14
|
+
* every assertion that did not pass.
|
|
15
|
+
*
|
|
16
|
+
* @example
|
|
17
|
+
* ```ts
|
|
18
|
+
* import { describeEval } from '@cogitator-ai/evals/vitest';
|
|
19
|
+
*
|
|
20
|
+
* const qa = describeEval('qa agent', { suite, run: { maxCases: 50 } });
|
|
21
|
+
*
|
|
22
|
+
* afterAll(() => qa.result().report('junit', { path: './reports/qa' }));
|
|
23
|
+
* ```
|
|
24
|
+
*/
|
|
25
|
+
export function describeEval(name, options) {
|
|
26
|
+
const { suite, run = {}, timeout = DEFAULT_TIMEOUT } = options;
|
|
27
|
+
if (!Number.isFinite(timeout) || timeout <= 0 || timeout > MAX_TIMEOUT) {
|
|
28
|
+
throw new Error(`describeEval timeout must be a positive number of milliseconds up to ${MAX_TIMEOUT}, got ${timeout}`);
|
|
29
|
+
}
|
|
30
|
+
let outcome = { state: 'pending' };
|
|
31
|
+
const result = () => {
|
|
32
|
+
switch (outcome.state) {
|
|
33
|
+
case 'done':
|
|
34
|
+
return outcome.result;
|
|
35
|
+
case 'failed':
|
|
36
|
+
throw new Error(`Eval suite '${name}' failed to run: ${messageOf(outcome.error)}`, {
|
|
37
|
+
cause: outcome.error,
|
|
38
|
+
});
|
|
39
|
+
case 'pending':
|
|
40
|
+
throw new Error(`Eval suite '${name}' has not finished running: read the result inside its tests or afterAll`);
|
|
41
|
+
}
|
|
42
|
+
};
|
|
43
|
+
describe(name, () => {
|
|
44
|
+
beforeAll(async () => {
|
|
45
|
+
const deadline = new AbortController();
|
|
46
|
+
const late = new Error(`the run did not finish within ${timeout} ms`);
|
|
47
|
+
let timer;
|
|
48
|
+
const expired = new Promise((_resolve, reject) => {
|
|
49
|
+
timer = setTimeout(() => {
|
|
50
|
+
deadline.abort(late);
|
|
51
|
+
reject(late);
|
|
52
|
+
}, timeout);
|
|
53
|
+
});
|
|
54
|
+
const signal = run.signal ? AbortSignal.any([run.signal, deadline.signal]) : deadline.signal;
|
|
55
|
+
const running = suite.run({ ...run, signal });
|
|
56
|
+
running.catch(() => undefined);
|
|
57
|
+
try {
|
|
58
|
+
outcome = { state: 'done', result: await Promise.race([running, expired]) };
|
|
59
|
+
}
|
|
60
|
+
catch (error) {
|
|
61
|
+
outcome = { state: 'failed', error };
|
|
62
|
+
}
|
|
63
|
+
finally {
|
|
64
|
+
clearTimeout(timer);
|
|
65
|
+
}
|
|
66
|
+
}, timeout + HOOK_GRACE);
|
|
67
|
+
suite.cases(run.maxCases).forEach((evalCase, index) => {
|
|
68
|
+
it(caseLabel(index, evalCase.input), () => {
|
|
69
|
+
const caseResult = result().results[index];
|
|
70
|
+
if (caseResult?.error !== undefined) {
|
|
71
|
+
const attempts = caseResult.attempts === undefined ? '' : ` after ${caseResult.attempts} attempt(s)`;
|
|
72
|
+
throw new Error(`Case failed${attempts}: ${caseResult.error}`);
|
|
73
|
+
}
|
|
74
|
+
});
|
|
75
|
+
});
|
|
76
|
+
it('assertions', () => {
|
|
77
|
+
const failed = result().assertions.filter((a) => !a.passed);
|
|
78
|
+
if (failed.length > 0) {
|
|
79
|
+
throw new Error(`${failed.length} assertion(s) failed:\n${failed.map((a) => ` ${a.name}: ${a.message}`).join('\n')}`);
|
|
80
|
+
}
|
|
81
|
+
});
|
|
82
|
+
});
|
|
83
|
+
return { result };
|
|
84
|
+
}
|
|
85
|
+
//# sourceMappingURL=vitest.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"vitest.js","sourceRoot":"","sources":["../src/vitest.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,SAAS,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,QAAQ,CAAC;AAEjD,OAAO,EAAE,SAAS,EAAE,MAAM,oBAAoB,CAAC;AAE/C,MAAM,eAAe,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC;AACvC,oFAAoF;AACpF,MAAM,UAAU,GAAG,CAAC,GAAG,IAAI,CAAC;AAC5B,+EAA+E;AAC/E,MAAM,WAAW,GAAG,aAAa,GAAG,UAAU,CAAC;AA2B/C,SAAS,SAAS,CAAC,KAAc;IAC/B,OAAO,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;AAChE,CAAC;AAED;;;;;;;;;;;;;GAaG;AACH,MAAM,UAAU,YAAY,CAAC,IAAY,EAAE,OAA4B;IACrE,MAAM,EAAE,KAAK,EAAE,GAAG,GAAG,EAAE,EAAE,OAAO,GAAG,eAAe,EAAE,GAAG,OAAO,CAAC;IAC/D,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,OAAO,CAAC,IAAI,OAAO,IAAI,CAAC,IAAI,OAAO,GAAG,WAAW,EAAE,CAAC;QACvE,MAAM,IAAI,KAAK,CACb,wEAAwE,WAAW,SAAS,OAAO,EAAE,CACtG,CAAC;IACJ,CAAC;IAED,IAAI,OAAO,GAAY,EAAE,KAAK,EAAE,SAAS,EAAE,CAAC;IAE5C,MAAM,MAAM,GAAG,GAAoB,EAAE;QACnC,QAAQ,OAAO,CAAC,KAAK,EAAE,CAAC;YACtB,KAAK,MAAM;gBACT,OAAO,OAAO,CAAC,MAAM,CAAC;YACxB,KAAK,QAAQ;gBACX,MAAM,IAAI,KAAK,CAAC,eAAe,IAAI,oBAAoB,SAAS,CAAC,OAAO,CAAC,KAAK,CAAC,EAAE,EAAE;oBACjF,KAAK,EAAE,OAAO,CAAC,KAAK;iBACrB,CAAC,CAAC;YACL,KAAK,SAAS;gBACZ,MAAM,IAAI,KAAK,CACb,eAAe,IAAI,0EAA0E,CAC9F,CAAC;QACN,CAAC;IACH,CAAC,CAAC;IAEF,QAAQ,CAAC,IAAI,EAAE,GAAG,EAAE;QAClB,SAAS,CAAC,KAAK,IAAI,EAAE;YACnB,MAAM,QAAQ,GAAG,IAAI,eAAe,EAAE,CAAC;YACvC,MAAM,IAAI,GAAG,IAAI,KAAK,CAAC,iCAAiC,OAAO,KAAK,CAAC,CAAC;YACtE,IAAI,KAAgD,CAAC;YACrD,MAAM,OAAO,GAAG,IAAI,OAAO,CAAQ,CAAC,QAAQ,EAAE,MAAM,EAAE,EAAE;gBACtD,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE;oBACtB,QAAQ,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;oBACrB,MAAM,CAAC,IAAI,CAAC,CAAC;gBACf,CAAC,EAAE,OAAO,CAAC,CAAC;YACd,CAAC,CAAC,CAAC;YACH,MAAM,MAAM,GAAG,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,MAAM,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC;YAC7F,MAAM,OAAO,GAAG,KAAK,CAAC,GAAG,CAAC,EAAE,GAAG,GAAG,EAAE,MAAM,EAAE,CAAC,CAAC;YAC9C,OAAO,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,SAAS,CAAC,CAAC;YAC/B,IAAI,CAAC;gBACH,OAAO,GAAG,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,OAAO,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC,EAAE,CAAC;YAC9E,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,OAAO,GAAG,EAAE,KAAK,EAAE,QAAQ,EAAE,KAAK,EAAE,CAAC;YACvC,CAAC;oBAAS,CAAC;gBACT,YAAY,CAAC,KAAK,CAAC,CAAC;YACtB,CAAC;QACH,CAAC,EAAE,OAAO,GAAG,UAAU,CAAC,CAAC;QAEzB,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,OAAO,CAAC,CAAC,QAAQ,EAAE,KAAK,EAAE,EAAE;YACpD,EAAE,CAAC,SAAS,CAAC,KAAK,EAAE,QAAQ,CAAC,KAAK,CAAC,EAAE,GAAG,EAAE;gBACxC,MAAM,UAAU,GAAG,MAAM,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;gBAC3C,IAAI,UAAU,EAAE,KAAK,KAAK,SAAS,EAAE,CAAC;oBACpC,MAAM,QAAQ,GACZ,UAAU,CAAC,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,UAAU,UAAU,CAAC,QAAQ,aAAa,CAAC;oBACtF,MAAM,IAAI,KAAK,CAAC,cAAc,QAAQ,KAAK,UAAU,CAAC,KAAK,EAAE,CAAC,CAAC;gBACjE,CAAC;YACH,CAAC,CAAC,CAAC;QACL,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,YAAY,EAAE,GAAG,EAAE;YACpB,MAAM,MAAM,GAAG,MAAM,EAAE,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC;YAC5D,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBACtB,MAAM,IAAI,KAAK,CACb,GAAG,MAAM,CAAC,MAAM,0BAA0B,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,KAAK,CAAC,CAAC,IAAI,KAAK,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CACtG,CAAC;YACJ,CAAC;QACH,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IAEH,OAAO,EAAE,MAAM,EAAE,CAAC;AACpB,CAAC"}
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@cogitator-ai/evals",
|
|
3
|
-
"version": "0.
|
|
3
|
+
"version": "0.10.0",
|
|
4
4
|
"description": "Evaluation framework for Cogitator AI agents",
|
|
5
5
|
"keywords": [
|
|
6
6
|
"cogitator",
|
|
@@ -21,6 +21,16 @@
|
|
|
21
21
|
"types": "./dist/index.d.ts",
|
|
22
22
|
"import": "./dist/index.js",
|
|
23
23
|
"default": "./dist/index.js"
|
|
24
|
+
},
|
|
25
|
+
"./vitest": {
|
|
26
|
+
"types": "./dist/vitest.d.ts",
|
|
27
|
+
"import": "./dist/vitest.js",
|
|
28
|
+
"default": "./dist/vitest.js"
|
|
29
|
+
},
|
|
30
|
+
"./matchers": {
|
|
31
|
+
"types": "./dist/matchers.d.ts",
|
|
32
|
+
"import": "./dist/matchers.js",
|
|
33
|
+
"default": "./dist/matchers.js"
|
|
24
34
|
}
|
|
25
35
|
},
|
|
26
36
|
"sideEffects": false,
|
|
@@ -34,18 +44,22 @@
|
|
|
34
44
|
"zod": "^4.6.5"
|
|
35
45
|
},
|
|
36
46
|
"peerDependencies": {
|
|
37
|
-
"@cogitator-ai/core": ">=0.
|
|
47
|
+
"@cogitator-ai/core": ">=0.58.0",
|
|
48
|
+
"vitest": ">=3.0.0"
|
|
38
49
|
},
|
|
39
50
|
"peerDependenciesMeta": {
|
|
40
51
|
"@cogitator-ai/core": {
|
|
41
52
|
"optional": true
|
|
53
|
+
},
|
|
54
|
+
"vitest": {
|
|
55
|
+
"optional": true
|
|
42
56
|
}
|
|
43
57
|
},
|
|
44
58
|
"optionalDependencies": {
|
|
45
59
|
"papaparse": "^5.7.0"
|
|
46
60
|
},
|
|
47
61
|
"devDependencies": {
|
|
48
|
-
"@cogitator-ai/core": "0.
|
|
62
|
+
"@cogitator-ai/core": "0.58.0",
|
|
49
63
|
"@types/papaparse": "^5.3.15",
|
|
50
64
|
"typescript": "^6.0.3",
|
|
51
65
|
"vitest": "^5.0.3"
|