@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.cjs
CHANGED
|
@@ -1731,6 +1731,7 @@ function processOneEvaluation(task, unit, totalEvaluations, publishEvent, persis
|
|
|
1731
1731
|
output,
|
|
1732
1732
|
meta: {
|
|
1733
1733
|
triggerId: task.triggerId,
|
|
1734
|
+
triggerTimestamp: task.triggerTimestamp,
|
|
1734
1735
|
runId: evaluatorRunId,
|
|
1735
1736
|
datasetName: task.dataset.getDisplayLabel(),
|
|
1736
1737
|
testCaseId: testCaseItem.id,
|
|
@@ -2211,6 +2212,7 @@ var EffectRunner = class {
|
|
|
2211
2212
|
const globalConcurrency = Math.max(1, request.globalConcurrency);
|
|
2212
2213
|
const sem = effect.Effect.unsafeMakeSemaphore(globalConcurrency);
|
|
2213
2214
|
const triggerId = request.triggerId ?? `trg-${crypto.randomUUID()}`;
|
|
2215
|
+
const triggerTimestamp = request.triggerTimestamp ?? Date.now();
|
|
2214
2216
|
const snapshots = [];
|
|
2215
2217
|
for (const job of request.jobs) {
|
|
2216
2218
|
snapshots.push(
|
|
@@ -2218,6 +2220,7 @@ var EffectRunner = class {
|
|
|
2218
2220
|
datasetId: job.datasetId,
|
|
2219
2221
|
evaluatorIds: job.evaluatorIds,
|
|
2220
2222
|
triggerId,
|
|
2223
|
+
triggerTimestamp,
|
|
2221
2224
|
maxConcurrency: this.config.maxConcurrency ?? 1,
|
|
2222
2225
|
globalEvaluationSemaphore: sem,
|
|
2223
2226
|
runConfigName: job.runConfigName,
|
|
@@ -2255,6 +2258,7 @@ var EffectRunner = class {
|
|
|
2255
2258
|
datasetId: request.datasetId,
|
|
2256
2259
|
evaluatorIds: request.evaluatorIds,
|
|
2257
2260
|
triggerId: request.triggerId,
|
|
2261
|
+
triggerTimestamp: request.triggerTimestamp ?? Date.now(),
|
|
2258
2262
|
maxConcurrency: request.concurrency ?? this.config.maxConcurrency ?? 1,
|
|
2259
2263
|
repetitions: request.repetitions,
|
|
2260
2264
|
runConfigName,
|
|
@@ -2282,6 +2286,7 @@ var EffectRunner = class {
|
|
|
2282
2286
|
const totalEvaluations = selectedTestCases.length * repetitions;
|
|
2283
2287
|
const runConfigTags = [...params.runConfigTags ?? []];
|
|
2284
2288
|
const triggerId = params.triggerId ?? `trg-${crypto.randomUUID()}`;
|
|
2289
|
+
const triggerTimestamp = params.triggerTimestamp ?? Date.now();
|
|
2285
2290
|
const runId = `run-${crypto.randomUUID()}`;
|
|
2286
2291
|
const artifactPath = createArtifactPath(this.config.artifactDirectory, params.datasetId, runId);
|
|
2287
2292
|
const snapshot = {
|
|
@@ -2325,6 +2330,7 @@ var EffectRunner = class {
|
|
|
2325
2330
|
effect.Queue.offer(this.runQueue, {
|
|
2326
2331
|
runId,
|
|
2327
2332
|
triggerId,
|
|
2333
|
+
triggerTimestamp,
|
|
2328
2334
|
datasetId: params.datasetId,
|
|
2329
2335
|
dataset: dataset.dataset,
|
|
2330
2336
|
evaluators: selectedEvaluators,
|
|
@@ -3103,7 +3109,8 @@ function EvalsCliApp({ data, args, runner }) {
|
|
|
3103
3109
|
void runner.runDatasetWith({
|
|
3104
3110
|
datasetId: selectedDataset.id,
|
|
3105
3111
|
evaluatorIds: clampedState.selectedEvaluatorIds,
|
|
3106
|
-
...PROGRAMMATIC_RUN_CONFIG
|
|
3112
|
+
...PROGRAMMATIC_RUN_CONFIG,
|
|
3113
|
+
triggerTimestamp: Date.now()
|
|
3107
3114
|
}).then((snapshot) => {
|
|
3108
3115
|
setRuntimeMessage(
|
|
3109
3116
|
`Started ${snapshot.runId} on ${selectedDataset.name} (${snapshot.totalTestCases} cases).`
|