@m4trix/evals 0.29.0 → 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 +18 -6
- package/dist/cli-simple.cjs.map +1 -1
- package/dist/cli-simple.js +18 -6
- package/dist/cli-simple.js.map +1 -1
- package/dist/cli.cjs +8 -1
- package/dist/cli.cjs.map +1 -1
- package/dist/cli.js +8 -1
- package/dist/cli.js.map +1 -1
- package/dist/index.cjs +6 -0
- package/dist/index.cjs.map +1 -1
- package/dist/index.d.ts +13 -0
- package/dist/index.js +6 -0
- package/dist/index.js.map +1 -1
- package/package.json +1 -1
package/dist/cli.js
CHANGED
|
@@ -1704,6 +1704,7 @@ function processOneEvaluation(task, unit, totalEvaluations, publishEvent, persis
|
|
|
1704
1704
|
output,
|
|
1705
1705
|
meta: {
|
|
1706
1706
|
triggerId: task.triggerId,
|
|
1707
|
+
triggerTimestamp: task.triggerTimestamp,
|
|
1707
1708
|
runId: evaluatorRunId,
|
|
1708
1709
|
datasetName: task.dataset.getDisplayLabel(),
|
|
1709
1710
|
testCaseId: testCaseItem.id,
|
|
@@ -2184,6 +2185,7 @@ var EffectRunner = class {
|
|
|
2184
2185
|
const globalConcurrency = Math.max(1, request.globalConcurrency);
|
|
2185
2186
|
const sem = Effect.unsafeMakeSemaphore(globalConcurrency);
|
|
2186
2187
|
const triggerId = request.triggerId ?? `trg-${randomUUID()}`;
|
|
2188
|
+
const triggerTimestamp = request.triggerTimestamp ?? Date.now();
|
|
2187
2189
|
const snapshots = [];
|
|
2188
2190
|
for (const job of request.jobs) {
|
|
2189
2191
|
snapshots.push(
|
|
@@ -2191,6 +2193,7 @@ var EffectRunner = class {
|
|
|
2191
2193
|
datasetId: job.datasetId,
|
|
2192
2194
|
evaluatorIds: job.evaluatorIds,
|
|
2193
2195
|
triggerId,
|
|
2196
|
+
triggerTimestamp,
|
|
2194
2197
|
maxConcurrency: this.config.maxConcurrency ?? 1,
|
|
2195
2198
|
globalEvaluationSemaphore: sem,
|
|
2196
2199
|
runConfigName: job.runConfigName,
|
|
@@ -2228,6 +2231,7 @@ var EffectRunner = class {
|
|
|
2228
2231
|
datasetId: request.datasetId,
|
|
2229
2232
|
evaluatorIds: request.evaluatorIds,
|
|
2230
2233
|
triggerId: request.triggerId,
|
|
2234
|
+
triggerTimestamp: request.triggerTimestamp ?? Date.now(),
|
|
2231
2235
|
maxConcurrency: request.concurrency ?? this.config.maxConcurrency ?? 1,
|
|
2232
2236
|
repetitions: request.repetitions,
|
|
2233
2237
|
runConfigName,
|
|
@@ -2255,6 +2259,7 @@ var EffectRunner = class {
|
|
|
2255
2259
|
const totalEvaluations = selectedTestCases.length * repetitions;
|
|
2256
2260
|
const runConfigTags = [...params.runConfigTags ?? []];
|
|
2257
2261
|
const triggerId = params.triggerId ?? `trg-${randomUUID()}`;
|
|
2262
|
+
const triggerTimestamp = params.triggerTimestamp ?? Date.now();
|
|
2258
2263
|
const runId = `run-${randomUUID()}`;
|
|
2259
2264
|
const artifactPath = createArtifactPath(this.config.artifactDirectory, params.datasetId, runId);
|
|
2260
2265
|
const snapshot = {
|
|
@@ -2298,6 +2303,7 @@ var EffectRunner = class {
|
|
|
2298
2303
|
Queue.offer(this.runQueue, {
|
|
2299
2304
|
runId,
|
|
2300
2305
|
triggerId,
|
|
2306
|
+
triggerTimestamp,
|
|
2301
2307
|
datasetId: params.datasetId,
|
|
2302
2308
|
dataset: dataset.dataset,
|
|
2303
2309
|
evaluators: selectedEvaluators,
|
|
@@ -3076,7 +3082,8 @@ function EvalsCliApp({ data, args, runner }) {
|
|
|
3076
3082
|
void runner.runDatasetWith({
|
|
3077
3083
|
datasetId: selectedDataset.id,
|
|
3078
3084
|
evaluatorIds: clampedState.selectedEvaluatorIds,
|
|
3079
|
-
...PROGRAMMATIC_RUN_CONFIG
|
|
3085
|
+
...PROGRAMMATIC_RUN_CONFIG,
|
|
3086
|
+
triggerTimestamp: Date.now()
|
|
3080
3087
|
}).then((snapshot) => {
|
|
3081
3088
|
setRuntimeMessage(
|
|
3082
3089
|
`Started ${snapshot.runId} on ${selectedDataset.name} (${snapshot.totalTestCases} cases).`
|