@m4trix/evals 0.20.0 → 0.21.1
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/dist/cli-simple.cjs +222 -28
- package/dist/cli-simple.cjs.map +1 -1
- package/dist/cli-simple.js +221 -27
- package/dist/cli-simple.js.map +1 -1
- package/dist/cli.cjs +115 -5
- package/dist/cli.cjs.map +1 -1
- package/dist/cli.js +114 -5
- package/dist/cli.js.map +1 -1
- package/dist/index.cjs +117 -5
- package/dist/index.cjs.map +1 -1
- package/dist/index.d.ts +10 -2
- package/dist/index.js +114 -5
- package/dist/index.js.map +1 -1
- package/package.json +3 -3
package/dist/cli.cjs
CHANGED
|
@@ -13,7 +13,8 @@ var fs = require('fs');
|
|
|
13
13
|
var jitiModule = require('jiti');
|
|
14
14
|
var promises = require('fs/promises');
|
|
15
15
|
var url = require('url');
|
|
16
|
-
var
|
|
16
|
+
var diff = require('diff');
|
|
17
|
+
var stringify = require('fast-json-stable-stringify');
|
|
17
18
|
|
|
18
19
|
var _documentCurrentScript = typeof document !== 'undefined' ? document.currentScript : null;
|
|
19
20
|
function _interopDefault (e) { return e && e.__esModule ? e : { default: e }; }
|
|
@@ -38,6 +39,7 @@ function _interopNamespace(e) {
|
|
|
38
39
|
|
|
39
40
|
var React2__default = /*#__PURE__*/_interopDefault(React2);
|
|
40
41
|
var jitiModule__namespace = /*#__PURE__*/_interopNamespace(jitiModule);
|
|
42
|
+
var stringify__default = /*#__PURE__*/_interopDefault(stringify);
|
|
41
43
|
|
|
42
44
|
var SEP = " ";
|
|
43
45
|
var ARROW = "\u203A";
|
|
@@ -1004,10 +1006,102 @@ async function collectTestCasesFromFiles(config) {
|
|
|
1004
1006
|
);
|
|
1005
1007
|
return found.flat();
|
|
1006
1008
|
}
|
|
1009
|
+
function preprocessForDiff(value, options) {
|
|
1010
|
+
if (options?.sort && Array.isArray(value)) {
|
|
1011
|
+
return [...value].sort((a, b) => {
|
|
1012
|
+
const aStr = stringify__default.default(preprocessForDiff(a, options));
|
|
1013
|
+
const bStr = stringify__default.default(preprocessForDiff(b, options));
|
|
1014
|
+
return aStr.localeCompare(bStr);
|
|
1015
|
+
}).map((item) => preprocessForDiff(item, options));
|
|
1016
|
+
}
|
|
1017
|
+
if (value !== null && typeof value === "object" && !Array.isArray(value) && options?.excludeKeys) {
|
|
1018
|
+
const keys = Array.isArray(options.excludeKeys) ? options.excludeKeys : options.excludeKeys.split(",").map((k) => k.trim());
|
|
1019
|
+
const filtered = {};
|
|
1020
|
+
for (const [k, v] of Object.entries(value)) {
|
|
1021
|
+
if (!keys.includes(k)) {
|
|
1022
|
+
filtered[k] = preprocessForDiff(v, options);
|
|
1023
|
+
}
|
|
1024
|
+
}
|
|
1025
|
+
return filtered;
|
|
1026
|
+
}
|
|
1027
|
+
if (value !== null && typeof value === "object" && !Array.isArray(value)) {
|
|
1028
|
+
const result = {};
|
|
1029
|
+
for (const [k, v] of Object.entries(value)) {
|
|
1030
|
+
result[k] = preprocessForDiff(v, options);
|
|
1031
|
+
}
|
|
1032
|
+
return result;
|
|
1033
|
+
}
|
|
1034
|
+
if (typeof value === "number" && options?.precision !== void 0) {
|
|
1035
|
+
return Number(value.toFixed(options.precision));
|
|
1036
|
+
}
|
|
1037
|
+
return value;
|
|
1038
|
+
}
|
|
1039
|
+
function toPrettyJson(value) {
|
|
1040
|
+
const str = stringify__default.default(value);
|
|
1041
|
+
try {
|
|
1042
|
+
const parsed = JSON.parse(str);
|
|
1043
|
+
return JSON.stringify(parsed, null, 2);
|
|
1044
|
+
} catch {
|
|
1045
|
+
return str;
|
|
1046
|
+
}
|
|
1047
|
+
}
|
|
1048
|
+
function formatDiffParts(parts) {
|
|
1049
|
+
const lines = [];
|
|
1050
|
+
for (const part of parts) {
|
|
1051
|
+
const prefix = part.added ? "+ " : part.removed ? "- " : "";
|
|
1052
|
+
const partLines = part.value.split("\n");
|
|
1053
|
+
for (let i = 0; i < partLines.length; i++) {
|
|
1054
|
+
const line = partLines[i];
|
|
1055
|
+
if (i === partLines.length - 1 && line === "")
|
|
1056
|
+
continue;
|
|
1057
|
+
lines.push(prefix + line);
|
|
1058
|
+
}
|
|
1059
|
+
}
|
|
1060
|
+
return lines.join("\n");
|
|
1061
|
+
}
|
|
1007
1062
|
function createDiffString(expected, actual, diffOptions) {
|
|
1008
|
-
const
|
|
1009
|
-
const
|
|
1010
|
-
|
|
1063
|
+
const expectedProcessed = preprocessForDiff(expected, diffOptions);
|
|
1064
|
+
const actualProcessed = preprocessForDiff(actual, diffOptions);
|
|
1065
|
+
if (diffOptions?.keysOnly) {
|
|
1066
|
+
const expectedKeys = JSON.stringify(
|
|
1067
|
+
extractKeys(expectedProcessed),
|
|
1068
|
+
null,
|
|
1069
|
+
2
|
|
1070
|
+
);
|
|
1071
|
+
const actualKeys = JSON.stringify(
|
|
1072
|
+
extractKeys(actualProcessed),
|
|
1073
|
+
null,
|
|
1074
|
+
2
|
|
1075
|
+
);
|
|
1076
|
+
const parts2 = diff.diffLines(expectedKeys, actualKeys);
|
|
1077
|
+
return formatDiffParts(parts2);
|
|
1078
|
+
}
|
|
1079
|
+
const expectedStr = toPrettyJson(expectedProcessed);
|
|
1080
|
+
const actualStr = toPrettyJson(actualProcessed);
|
|
1081
|
+
if (expectedStr === actualStr) {
|
|
1082
|
+
return "";
|
|
1083
|
+
}
|
|
1084
|
+
const parts = diff.diffLines(expectedStr, actualStr);
|
|
1085
|
+
if (diffOptions?.outputNewOnly) {
|
|
1086
|
+
const filtered = parts.filter(
|
|
1087
|
+
(p) => p.added === true
|
|
1088
|
+
);
|
|
1089
|
+
return formatDiffParts(filtered);
|
|
1090
|
+
}
|
|
1091
|
+
return formatDiffParts(parts);
|
|
1092
|
+
}
|
|
1093
|
+
function extractKeys(value) {
|
|
1094
|
+
if (value === null || typeof value !== "object") {
|
|
1095
|
+
return "\xB7";
|
|
1096
|
+
}
|
|
1097
|
+
if (Array.isArray(value)) {
|
|
1098
|
+
return value.map(extractKeys);
|
|
1099
|
+
}
|
|
1100
|
+
const result = {};
|
|
1101
|
+
for (const [k, v] of Object.entries(value)) {
|
|
1102
|
+
result[k] = extractKeys(v);
|
|
1103
|
+
}
|
|
1104
|
+
return result;
|
|
1011
1105
|
}
|
|
1012
1106
|
function formatLogMessage(msg) {
|
|
1013
1107
|
if (typeof msg === "string")
|
|
@@ -1358,13 +1452,27 @@ function createArtifactPath(artifactDirectory, datasetId, runId) {
|
|
|
1358
1452
|
`${datasetId}_${runId}_${nowIsoForFile()}.jsonl`
|
|
1359
1453
|
);
|
|
1360
1454
|
}
|
|
1361
|
-
function processOneTestCase(task, testCaseItem, totalEvaluations, publishEvent, persistenceQueue, updateSnapshot, completedRef, passedRef, failedRef) {
|
|
1455
|
+
function processOneTestCase(task, testCaseItem, totalEvaluations, publishEvent, persistenceQueue, updateSnapshot, startedRef, completedRef, passedRef, failedRef) {
|
|
1362
1456
|
return effect.Effect.gen(function* () {
|
|
1363
1457
|
const reruns = typeof testCaseItem.testCase.getReruns === "function" ? testCaseItem.testCase.getReruns() : 1;
|
|
1364
1458
|
const rerunPassed = [];
|
|
1365
1459
|
for (let r = 0; r < reruns; r++) {
|
|
1366
1460
|
const evaluatorRunId = `run-${crypto.randomUUID()}`;
|
|
1367
1461
|
const started = Date.now();
|
|
1462
|
+
const startedEvaluations = yield* effect.Ref.modify(startedRef, (n) => [
|
|
1463
|
+
n + 1,
|
|
1464
|
+
n + 1
|
|
1465
|
+
]);
|
|
1466
|
+
yield* publishEvent({
|
|
1467
|
+
type: "TestCaseStarted",
|
|
1468
|
+
runId: task.runId,
|
|
1469
|
+
testCaseId: testCaseItem.id,
|
|
1470
|
+
testCaseName: testCaseItem.testCase.getName(),
|
|
1471
|
+
startedTestCases: startedEvaluations,
|
|
1472
|
+
totalTestCases: totalEvaluations,
|
|
1473
|
+
rerunIndex: r + 1,
|
|
1474
|
+
rerunTotal: reruns
|
|
1475
|
+
});
|
|
1368
1476
|
const evaluatorScores = [];
|
|
1369
1477
|
let testCaseError;
|
|
1370
1478
|
const output = readOutput(testCaseItem.testCase);
|
|
@@ -1510,6 +1618,7 @@ var executeRunTask = (task, publishEvent, persistenceQueue, updateSnapshot) => e
|
|
|
1510
1618
|
);
|
|
1511
1619
|
const maxConcurrency = Math.max(1, task.maxConcurrency ?? 1);
|
|
1512
1620
|
const completedRef = yield* effect.Ref.make(0);
|
|
1621
|
+
const startedRef = yield* effect.Ref.make(0);
|
|
1513
1622
|
const passedRef = yield* effect.Ref.make(0);
|
|
1514
1623
|
const failedRef = yield* effect.Ref.make(0);
|
|
1515
1624
|
const processTestCase = (testCaseItem) => processOneTestCase(
|
|
@@ -1519,6 +1628,7 @@ var executeRunTask = (task, publishEvent, persistenceQueue, updateSnapshot) => e
|
|
|
1519
1628
|
publishEvent,
|
|
1520
1629
|
persistenceQueue,
|
|
1521
1630
|
updateSnapshot,
|
|
1631
|
+
startedRef,
|
|
1522
1632
|
completedRef,
|
|
1523
1633
|
passedRef,
|
|
1524
1634
|
failedRef
|