@m4trix/evals 0.28.0 → 0.30.0
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 +1 -1
- package/dist/cli-simple.cjs +20 -6
- package/dist/cli-simple.cjs.map +1 -1
- package/dist/cli-simple.js +20 -6
- package/dist/cli-simple.js.map +1 -1
- package/dist/cli.cjs +10 -1
- package/dist/cli.cjs.map +1 -1
- package/dist/cli.js +10 -1
- package/dist/cli.js.map +1 -1
- package/dist/index.cjs +8 -0
- package/dist/index.cjs.map +1 -1
- package/dist/index.d.ts +17 -0
- package/dist/index.js +8 -0
- package/dist/index.js.map +1 -1
- package/package.json +1 -1
package/dist/index.cjs
CHANGED
|
@@ -1591,8 +1591,11 @@ function processOneEvaluation(task, unit, totalEvaluations, publishEvent, persis
|
|
|
1591
1591
|
output,
|
|
1592
1592
|
meta: {
|
|
1593
1593
|
triggerId: task.triggerId,
|
|
1594
|
+
triggerTimestamp: task.triggerTimestamp,
|
|
1594
1595
|
runId: evaluatorRunId,
|
|
1595
1596
|
datasetName: task.dataset.getDisplayLabel(),
|
|
1597
|
+
testCaseId: testCaseItem.id,
|
|
1598
|
+
testCaseName: getTestCaseDisplayLabel(testCaseItem.testCase),
|
|
1596
1599
|
repetitionId,
|
|
1597
1600
|
repetitionIndex,
|
|
1598
1601
|
repetitionCount,
|
|
@@ -2069,6 +2072,7 @@ var EffectRunner = class {
|
|
|
2069
2072
|
const globalConcurrency = Math.max(1, request.globalConcurrency);
|
|
2070
2073
|
const sem = effect.Effect.unsafeMakeSemaphore(globalConcurrency);
|
|
2071
2074
|
const triggerId = request.triggerId ?? `trg-${crypto.randomUUID()}`;
|
|
2075
|
+
const triggerTimestamp = request.triggerTimestamp ?? Date.now();
|
|
2072
2076
|
const snapshots = [];
|
|
2073
2077
|
for (const job of request.jobs) {
|
|
2074
2078
|
snapshots.push(
|
|
@@ -2076,6 +2080,7 @@ var EffectRunner = class {
|
|
|
2076
2080
|
datasetId: job.datasetId,
|
|
2077
2081
|
evaluatorIds: job.evaluatorIds,
|
|
2078
2082
|
triggerId,
|
|
2083
|
+
triggerTimestamp,
|
|
2079
2084
|
maxConcurrency: this.config.maxConcurrency ?? 1,
|
|
2080
2085
|
globalEvaluationSemaphore: sem,
|
|
2081
2086
|
runConfigName: job.runConfigName,
|
|
@@ -2113,6 +2118,7 @@ var EffectRunner = class {
|
|
|
2113
2118
|
datasetId: request.datasetId,
|
|
2114
2119
|
evaluatorIds: request.evaluatorIds,
|
|
2115
2120
|
triggerId: request.triggerId,
|
|
2121
|
+
triggerTimestamp: request.triggerTimestamp ?? Date.now(),
|
|
2116
2122
|
maxConcurrency: request.concurrency ?? this.config.maxConcurrency ?? 1,
|
|
2117
2123
|
repetitions: request.repetitions,
|
|
2118
2124
|
runConfigName,
|
|
@@ -2140,6 +2146,7 @@ var EffectRunner = class {
|
|
|
2140
2146
|
const totalEvaluations = selectedTestCases.length * repetitions;
|
|
2141
2147
|
const runConfigTags = [...params.runConfigTags ?? []];
|
|
2142
2148
|
const triggerId = params.triggerId ?? `trg-${crypto.randomUUID()}`;
|
|
2149
|
+
const triggerTimestamp = params.triggerTimestamp ?? Date.now();
|
|
2143
2150
|
const runId = `run-${crypto.randomUUID()}`;
|
|
2144
2151
|
const artifactPath = createArtifactPath(this.config.artifactDirectory, params.datasetId, runId);
|
|
2145
2152
|
const snapshot = {
|
|
@@ -2183,6 +2190,7 @@ var EffectRunner = class {
|
|
|
2183
2190
|
effect.Queue.offer(this.runQueue, {
|
|
2184
2191
|
runId,
|
|
2185
2192
|
triggerId,
|
|
2193
|
+
triggerTimestamp,
|
|
2186
2194
|
datasetId: params.datasetId,
|
|
2187
2195
|
dataset: dataset.dataset,
|
|
2188
2196
|
evaluators: selectedEvaluators,
|