agentv 2.14.0-next.1 → 2.14.2

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.
@@ -11,7 +11,7 @@ import {
11
11
  validateEvalFile,
12
12
  validateFileReferences,
13
13
  validateTargetsFile
14
- } from "./chunk-K2APOWTE.js";
14
+ } from "./chunk-LHDGRK2S.js";
15
15
  import {
16
16
  RepoManager,
17
17
  assembleLlmJudgePrompt,
@@ -4087,7 +4087,7 @@ var evalRunCommand = command({
4087
4087
  },
4088
4088
  handler: async (args) => {
4089
4089
  if (args.evalPaths.length === 0 && process.stdin.isTTY) {
4090
- const { launchInteractiveWizard } = await import("./interactive-WF6UO63B.js");
4090
+ const { launchInteractiveWizard } = await import("./interactive-R22BRR6A.js");
4091
4091
  await launchInteractiveWizard();
4092
4092
  return;
4093
4093
  }
@@ -5793,4 +5793,4 @@ export {
5793
5793
  preprocessArgv,
5794
5794
  runCli
5795
5795
  };
5796
- //# sourceMappingURL=chunk-ZSSGXZX6.js.map
5796
+ //# sourceMappingURL=chunk-IO5LMHZH.js.map
@@ -30,7 +30,7 @@ import {
30
30
  // package.json
31
31
  var package_default = {
32
32
  name: "agentv",
33
- version: "2.14.0-next.1",
33
+ version: "2.14.2",
34
34
  description: "CLI entry point for AgentV",
35
35
  type: "module",
36
36
  repository: {
@@ -3178,39 +3178,44 @@ async function runEvalCommand(input) {
3178
3178
  if (!targetPrep) {
3179
3179
  throw new Error(`Missing metadata for ${testFilePath}`);
3180
3180
  }
3181
- for (const { selection, inlineTargetLabel } of targetPrep.selections) {
3182
- const targetName = selection.targetName;
3183
- const applicableEvalCases = targetPrep.selections.length > 1 ? targetPrep.evalCases.filter((test) => {
3184
- if (test.targets && test.targets.length > 0) {
3185
- return test.targets.includes(targetName);
3181
+ const targetResults = await Promise.all(
3182
+ targetPrep.selections.map(async ({ selection, inlineTargetLabel }) => {
3183
+ const targetName = selection.targetName;
3184
+ const applicableEvalCases = targetPrep.selections.length > 1 ? targetPrep.evalCases.filter((test) => {
3185
+ if (test.targets && test.targets.length > 0) {
3186
+ return test.targets.includes(targetName);
3187
+ }
3188
+ return true;
3189
+ }) : targetPrep.evalCases;
3190
+ if (applicableEvalCases.length === 0) {
3191
+ return [];
3186
3192
  }
3187
- return true;
3188
- }) : targetPrep.evalCases;
3189
- if (applicableEvalCases.length === 0) {
3190
- continue;
3191
- }
3192
- const result = await runSingleEvalFile({
3193
- testFilePath,
3194
- cwd,
3195
- repoRoot,
3196
- options,
3197
- outputWriter,
3198
- otelExporter,
3199
- cache,
3200
- evaluationRunner,
3201
- workersOverride: perFileWorkers,
3202
- progressReporter,
3203
- seenEvalCases,
3204
- displayIdTracker,
3205
- selection,
3206
- inlineTargetLabel,
3207
- evalCases: applicableEvalCases,
3208
- trialsConfig: targetPrep.trialsConfig,
3209
- matrixMode: targetPrep.selections.length > 1,
3210
- totalBudgetUsd: targetPrep.totalBudgetUsd,
3211
- failOnError: targetPrep.failOnError
3212
- });
3213
- allResults.push(...result.results);
3193
+ const result = await runSingleEvalFile({
3194
+ testFilePath,
3195
+ cwd,
3196
+ repoRoot,
3197
+ options,
3198
+ outputWriter,
3199
+ otelExporter,
3200
+ cache,
3201
+ evaluationRunner,
3202
+ workersOverride: perFileWorkers,
3203
+ progressReporter,
3204
+ seenEvalCases,
3205
+ displayIdTracker,
3206
+ selection,
3207
+ inlineTargetLabel,
3208
+ evalCases: applicableEvalCases,
3209
+ trialsConfig: targetPrep.trialsConfig,
3210
+ matrixMode: targetPrep.selections.length > 1,
3211
+ totalBudgetUsd: targetPrep.totalBudgetUsd,
3212
+ failOnError: targetPrep.failOnError
3213
+ });
3214
+ return result.results;
3215
+ })
3216
+ );
3217
+ for (const results of targetResults) {
3218
+ allResults.push(...results);
3214
3219
  }
3215
3220
  });
3216
3221
  progressReporter.finish();
@@ -3294,4 +3299,4 @@ export {
3294
3299
  selectTarget,
3295
3300
  runEvalCommand
3296
3301
  };
3297
- //# sourceMappingURL=chunk-K2APOWTE.js.map
3302
+ //# sourceMappingURL=chunk-LHDGRK2S.js.map