agentv 4.2.0 → 4.3.0
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/dist/{chunk-UXSQQHCI.js → chunk-ASU5L5ZW.js} +15 -27
- package/dist/chunk-ASU5L5ZW.js.map +1 -0
- package/dist/{chunk-HAZJO7OY.js → chunk-ZDJN5FSI.js} +2 -2
- package/dist/{chunk-HAZJO7OY.js.map → chunk-ZDJN5FSI.js.map} +1 -1
- package/dist/cli.js +2 -2
- package/dist/index.js +2 -2
- package/dist/{interactive-NVNOLL2H.js → interactive-BKK53ETJ.js} +2 -2
- package/package.json +1 -1
- package/dist/chunk-UXSQQHCI.js.map +0 -1
- /package/dist/{interactive-NVNOLL2H.js.map → interactive-BKK53ETJ.js.map} +0 -0
|
@@ -24,7 +24,7 @@ import {
|
|
|
24
24
|
validateFileReferences,
|
|
25
25
|
validateTargetsFile,
|
|
26
26
|
writeArtifactsFromResults
|
|
27
|
-
} from "./chunk-
|
|
27
|
+
} from "./chunk-ZDJN5FSI.js";
|
|
28
28
|
import {
|
|
29
29
|
DEFAULT_CATEGORY,
|
|
30
30
|
createBuiltinRegistry,
|
|
@@ -4217,7 +4217,7 @@ var evalRunCommand = command({
|
|
|
4217
4217
|
},
|
|
4218
4218
|
handler: async (args) => {
|
|
4219
4219
|
if (args.evalPaths.length === 0 && process.stdin.isTTY) {
|
|
4220
|
-
const { launchInteractiveWizard } = await import("./interactive-
|
|
4220
|
+
const { launchInteractiveWizard } = await import("./interactive-BKK53ETJ.js");
|
|
4221
4221
|
await launchInteractiveWizard();
|
|
4222
4222
|
return;
|
|
4223
4223
|
}
|
|
@@ -4441,27 +4441,15 @@ var evalBenchCommand = command({
|
|
|
4441
4441
|
type: string,
|
|
4442
4442
|
displayName: "export-dir",
|
|
4443
4443
|
description: "Export directory from pipeline input/grade"
|
|
4444
|
-
}),
|
|
4445
|
-
llmScores: option({
|
|
4446
|
-
type: optional(string),
|
|
4447
|
-
long: "llm-scores",
|
|
4448
|
-
description: "Path to LLM scores JSON file (reads from stdin if omitted)"
|
|
4449
4444
|
})
|
|
4450
4445
|
},
|
|
4451
|
-
handler: async ({ exportDir
|
|
4446
|
+
handler: async ({ exportDir }) => {
|
|
4452
4447
|
const manifest = JSON.parse(await readFile(join(exportDir, "manifest.json"), "utf8"));
|
|
4453
4448
|
const testIds = manifest.test_ids;
|
|
4454
4449
|
const targetName = manifest.target?.name ?? "unknown";
|
|
4455
4450
|
const evalSet = manifest.dataset ?? "";
|
|
4456
4451
|
const experiment = manifest.experiment;
|
|
4457
4452
|
const safeEvalSet = evalSet ? evalSet.replace(/[\/\\:*?"<>|]/g, "_") : "";
|
|
4458
|
-
let stdinData;
|
|
4459
|
-
if (llmScoresPath) {
|
|
4460
|
-
stdinData = await readFile(llmScoresPath, "utf8");
|
|
4461
|
-
} else {
|
|
4462
|
-
stdinData = await readStdin();
|
|
4463
|
-
}
|
|
4464
|
-
const llmScores = stdinData ? JSON.parse(stdinData) : {};
|
|
4465
4453
|
const indexLines = [];
|
|
4466
4454
|
const allPassRates = [];
|
|
4467
4455
|
for (const testId of testIds) {
|
|
@@ -4488,14 +4476,18 @@ var evalBenchCommand = command({
|
|
|
4488
4476
|
}
|
|
4489
4477
|
} catch {
|
|
4490
4478
|
}
|
|
4491
|
-
const testLlmScores = llmScores[testId] ?? {};
|
|
4492
4479
|
const llmGradersDir = join(testDir, "llm_graders");
|
|
4493
4480
|
try {
|
|
4494
4481
|
const graderFiles = (await readdir(llmGradersDir)).filter((f) => f.endsWith(".json"));
|
|
4495
4482
|
for (const file of graderFiles) {
|
|
4496
4483
|
const graderMeta = JSON.parse(await readFile(join(llmGradersDir, file), "utf8"));
|
|
4497
4484
|
const graderName = graderMeta.name;
|
|
4498
|
-
const
|
|
4485
|
+
const diskResultPath = join(testDir, "llm_grader_results", `${graderName}.json`);
|
|
4486
|
+
let llmResult;
|
|
4487
|
+
try {
|
|
4488
|
+
llmResult = JSON.parse(await readFile(diskResultPath, "utf8"));
|
|
4489
|
+
} catch {
|
|
4490
|
+
}
|
|
4499
4491
|
if (llmResult) {
|
|
4500
4492
|
evaluators.push({
|
|
4501
4493
|
name: graderName,
|
|
@@ -4515,7 +4507,7 @@ var evalBenchCommand = command({
|
|
|
4515
4507
|
const weightedScore = totalWeight > 0 ? evaluators.reduce((sum, e) => sum + e.score * e.weight, 0) / totalWeight : 0;
|
|
4516
4508
|
const passed = allAssertions.filter((a) => a.passed).length;
|
|
4517
4509
|
const failed = allAssertions.filter((a) => !a.passed).length;
|
|
4518
|
-
const passRate = allAssertions.length > 0 ? Math.round(passed / allAssertions.length * 1e3) / 1e3 : 0;
|
|
4510
|
+
const passRate = allAssertions.length > 0 ? Math.round(passed / allAssertions.length * 1e3) / 1e3 : weightedScore >= 0.5 ? 1 : 0;
|
|
4519
4511
|
allPassRates.push(passRate);
|
|
4520
4512
|
const grading = {
|
|
4521
4513
|
assertions: allAssertions,
|
|
@@ -4608,13 +4600,6 @@ var evalBenchCommand = command({
|
|
|
4608
4600
|
console.log(`Benchmark: ${testIds.length} test(s), pass_rate=${passRateStats.mean}`);
|
|
4609
4601
|
}
|
|
4610
4602
|
});
|
|
4611
|
-
async function readStdin() {
|
|
4612
|
-
const chunks = [];
|
|
4613
|
-
for await (const chunk of process.stdin) {
|
|
4614
|
-
chunks.push(chunk);
|
|
4615
|
-
}
|
|
4616
|
-
return Buffer.concat(chunks).toString("utf8").trim();
|
|
4617
|
-
}
|
|
4618
4603
|
function computeStats(values) {
|
|
4619
4604
|
if (values.length === 0) return { mean: 0, stddev: 0 };
|
|
4620
4605
|
const mean2 = values.reduce((sum, v) => sum + v, 0) / values.length;
|
|
@@ -4681,7 +4666,10 @@ async function runCodeGraders(tasks, concurrency) {
|
|
|
4681
4666
|
);
|
|
4682
4667
|
const parsed = JSON.parse(stdout);
|
|
4683
4668
|
const score = typeof parsed.score === "number" ? parsed.score : 0;
|
|
4684
|
-
const assertions = Array.isArray(parsed.assertions) ? parsed.assertions : [
|
|
4669
|
+
const assertions = Array.isArray(parsed.assertions) && parsed.assertions.length > 0 ? parsed.assertions : [
|
|
4670
|
+
...(parsed.hits ?? []).map((h) => ({ text: h, passed: true })),
|
|
4671
|
+
...(parsed.misses ?? []).map((m) => ({ text: m, passed: false }))
|
|
4672
|
+
];
|
|
4685
4673
|
const result = {
|
|
4686
4674
|
name: graderName,
|
|
4687
4675
|
type: "code-grader",
|
|
@@ -8330,4 +8318,4 @@ export {
|
|
|
8330
8318
|
preprocessArgv,
|
|
8331
8319
|
runCli
|
|
8332
8320
|
};
|
|
8333
|
-
//# sourceMappingURL=chunk-
|
|
8321
|
+
//# sourceMappingURL=chunk-ASU5L5ZW.js.map
|