agent-hitch 0.2.7 → 0.2.9
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 +16 -2
- package/README.zh-CN.md +14 -2
- package/dist/bin/hitch.js +25 -3
- package/dist/bin/hitch.js.map +1 -1
- package/dist/scripts/canary-benchmark.js +54 -0
- package/dist/scripts/canary-benchmark.js.map +1 -0
- package/dist/scripts/canary-eval-scheduler-throughput.js +47 -0
- package/dist/scripts/canary-eval-scheduler-throughput.js.map +1 -0
- package/dist/scripts/check-architecture.js +3 -2
- package/dist/scripts/check-architecture.js.map +1 -1
- package/dist/src/adapters/catalog.js +2 -1
- package/dist/src/adapters/catalog.js.map +1 -1
- package/dist/src/adapters/providers/codex-auth.js +29 -0
- package/dist/src/adapters/providers/codex-auth.js.map +1 -0
- package/dist/src/adapters/providers/codex.js +23 -6
- package/dist/src/adapters/providers/codex.js.map +1 -1
- package/dist/src/adapters/providers/model-call.js +26 -0
- package/dist/src/adapters/providers/model-call.js.map +1 -0
- package/dist/src/artifacts/index.js +1 -1
- package/dist/src/artifacts/index.js.map +1 -1
- package/dist/src/artifacts/preparer.js +7 -2
- package/dist/src/artifacts/preparer.js.map +1 -1
- package/dist/src/backends/harbor/agent-budget.js +31 -0
- package/dist/src/backends/harbor/agent-budget.js.map +1 -0
- package/dist/src/backends/harbor/backend.js +6 -4
- package/dist/src/backends/harbor/backend.js.map +1 -1
- package/dist/src/backends/harbor/index.js +1 -0
- package/dist/src/backends/harbor/index.js.map +1 -1
- package/dist/src/backends/harbor/regrade.js +77 -0
- package/dist/src/backends/harbor/regrade.js.map +1 -0
- package/dist/src/backends/index.js +1 -0
- package/dist/src/backends/index.js.map +1 -1
- package/dist/src/benchmarks/index.js +3 -0
- package/dist/src/benchmarks/index.js.map +1 -0
- package/dist/src/benchmarks/loader.js +271 -0
- package/dist/src/benchmarks/loader.js.map +1 -0
- package/dist/src/benchmarks/metrics.js +16 -0
- package/dist/src/benchmarks/metrics.js.map +1 -0
- package/dist/src/benchmarks/toml.js +13 -0
- package/dist/src/benchmarks/toml.js.map +1 -0
- package/dist/src/benchmarks/validation.js +242 -0
- package/dist/src/benchmarks/validation.js.map +1 -0
- package/dist/src/cli/arguments.js +6 -4
- package/dist/src/cli/arguments.js.map +1 -1
- package/dist/src/cli/commands/benchmark.js +22 -0
- package/dist/src/cli/commands/benchmark.js.map +1 -0
- package/dist/src/cli/commands/capabilities.js +18 -0
- package/dist/src/cli/commands/capabilities.js.map +1 -0
- package/dist/src/cli/commands/eval.js +15 -3
- package/dist/src/cli/commands/eval.js.map +1 -1
- package/dist/src/cli/commands/run.js +37 -15
- package/dist/src/cli/commands/run.js.map +1 -1
- package/dist/src/cli/commands/trajectory.js +96 -5
- package/dist/src/cli/commands/trajectory.js.map +1 -1
- package/dist/src/cli/commands/verifier.js +36 -0
- package/dist/src/cli/commands/verifier.js.map +1 -0
- package/dist/src/cli/main.js +6 -0
- package/dist/src/cli/main.js.map +1 -1
- package/dist/src/cli/output.js +14 -2
- package/dist/src/cli/output.js.map +1 -1
- package/dist/src/control-plane/eval-control-work.js +5 -0
- package/dist/src/control-plane/eval-control-work.js.map +1 -1
- package/dist/src/control-plane/eval-scheduler.js +4 -1
- package/dist/src/control-plane/eval-scheduler.js.map +1 -1
- package/dist/src/control-plane/index.js +1 -1
- package/dist/src/control-plane/index.js.map +1 -1
- package/dist/src/control-plane/remote-result-transport.js +3 -0
- package/dist/src/control-plane/remote-result-transport.js.map +1 -1
- package/dist/src/control-plane/remote-work-coordinator.js +1 -0
- package/dist/src/control-plane/remote-work-coordinator.js.map +1 -1
- package/dist/src/control-plane/remote-work-item.js +62 -0
- package/dist/src/control-plane/remote-work-item.js.map +1 -0
- package/dist/src/control-plane/remote-work-recovery.js +44 -5
- package/dist/src/control-plane/remote-work-recovery.js.map +1 -1
- package/dist/src/control-plane/remote-worker-protocol.js +2 -23
- package/dist/src/control-plane/remote-worker-protocol.js.map +1 -1
- package/dist/src/control-plane/rerun-scheduler.js +51 -14
- package/dist/src/control-plane/rerun-scheduler.js.map +1 -1
- package/dist/src/control-plane/rerun-submission.js +10 -4
- package/dist/src/control-plane/rerun-submission.js.map +1 -1
- package/dist/src/control-plane/work-admission.js +2 -0
- package/dist/src/control-plane/work-admission.js.map +1 -1
- package/dist/src/control-plane/work-dispatcher.js +9 -1
- package/dist/src/control-plane/work-dispatcher.js.map +1 -1
- package/dist/src/controller-runtime/hash.js +7 -2
- package/dist/src/controller-runtime/hash.js.map +1 -1
- package/dist/src/daemon/auth.js +12 -2
- package/dist/src/daemon/auth.js.map +1 -1
- package/dist/src/domain/benchmarks.js +2 -0
- package/dist/src/domain/benchmarks.js.map +1 -0
- package/dist/src/domain/index.js +4 -0
- package/dist/src/domain/index.js.map +1 -1
- package/dist/src/domain/runs.js.map +1 -1
- package/dist/src/domain/validation.js +11 -4
- package/dist/src/domain/validation.js.map +1 -1
- package/dist/src/domain/verifier-evidence-validation.js +214 -0
- package/dist/src/domain/verifier-evidence-validation.js.map +1 -0
- package/dist/src/domain/verifier-evidence.js +3 -0
- package/dist/src/domain/verifier-evidence.js.map +1 -0
- package/dist/src/domain/verifier-score-contract.js +232 -0
- package/dist/src/domain/verifier-score-contract.js.map +1 -0
- package/dist/src/evals/benchmark-adapter-manifest.js +219 -0
- package/dist/src/evals/benchmark-adapter-manifest.js.map +1 -0
- package/dist/src/evals/benchmark-candidate.js +36 -0
- package/dist/src/evals/benchmark-candidate.js.map +1 -0
- package/dist/src/evals/benchmark-run.js +166 -0
- package/dist/src/evals/benchmark-run.js.map +1 -0
- package/dist/src/evals/collect-only-rerun.js +2 -1
- package/dist/src/evals/collect-only-rerun.js.map +1 -1
- package/dist/src/evals/directory.js +1 -1
- package/dist/src/evals/directory.js.map +1 -1
- package/dist/src/evals/duration-estimator.js +126 -0
- package/dist/src/evals/duration-estimator.js.map +1 -0
- package/dist/src/evals/eval-lifecycle-events.js +13 -0
- package/dist/src/evals/eval-lifecycle-events.js.map +1 -1
- package/dist/src/evals/eval-logical-plan.js +51 -0
- package/dist/src/evals/eval-logical-plan.js.map +1 -0
- package/dist/src/evals/eval-result-builder.js +81 -0
- package/dist/src/evals/eval-result-builder.js.map +1 -0
- package/dist/src/evals/evolution-baseline.js +191 -0
- package/dist/src/evals/evolution-baseline.js.map +1 -0
- package/dist/src/evals/execution-plan.js +39 -3
- package/dist/src/evals/execution-plan.js.map +1 -1
- package/dist/src/evals/failure-classifier.js +61 -0
- package/dist/src/evals/failure-classifier.js.map +1 -0
- package/dist/src/evals/harbor-artifact-builder.js +6 -2
- package/dist/src/evals/harbor-artifact-builder.js.map +1 -1
- package/dist/src/evals/harbor-bridge-error.js +42 -1
- package/dist/src/evals/harbor-bridge-error.js.map +1 -1
- package/dist/src/evals/harbor-node-runtime.js +146 -0
- package/dist/src/evals/harbor-node-runtime.js.map +1 -0
- package/dist/src/evals/index.js +15 -0
- package/dist/src/evals/index.js.map +1 -1
- package/dist/src/evals/infrastructure-retry.js +109 -22
- package/dist/src/evals/infrastructure-retry.js.map +1 -1
- package/dist/src/evals/native-phase-evidence.js +440 -0
- package/dist/src/evals/native-phase-evidence.js.map +1 -0
- package/dist/src/evals/physical-retry-work.js +31 -0
- package/dist/src/evals/physical-retry-work.js.map +1 -0
- package/dist/src/evals/planned-execution-support.js +22 -10
- package/dist/src/evals/planned-execution-support.js.map +1 -1
- package/dist/src/evals/planned-execution.js +120 -112
- package/dist/src/evals/planned-execution.js.map +1 -1
- package/dist/src/evals/planned-progress-publisher.js +62 -0
- package/dist/src/evals/planned-progress-publisher.js.map +1 -0
- package/dist/src/evals/planned-retry-execution.js +174 -0
- package/dist/src/evals/planned-retry-execution.js.map +1 -0
- package/dist/src/evals/planned-retry-lifecycle.js +4 -2
- package/dist/src/evals/planned-retry-lifecycle.js.map +1 -1
- package/dist/src/evals/preparation-rerun.js +225 -0
- package/dist/src/evals/preparation-rerun.js.map +1 -0
- package/dist/src/evals/progress.js +48 -8
- package/dist/src/evals/progress.js.map +1 -1
- package/dist/src/evals/recovery.js +54 -4
- package/dist/src/evals/recovery.js.map +1 -1
- package/dist/src/evals/regrade-evidence.js +77 -0
- package/dist/src/evals/regrade-evidence.js.map +1 -0
- package/dist/src/evals/remote-infrastructure-retry.js +84 -18
- package/dist/src/evals/remote-infrastructure-retry.js.map +1 -1
- package/dist/src/evals/request.js +21 -3
- package/dist/src/evals/request.js.map +1 -1
- package/dist/src/evals/rerun-inputs.js +73 -0
- package/dist/src/evals/rerun-inputs.js.map +1 -1
- package/dist/src/evals/rerun-types.js +2 -6
- package/dist/src/evals/rerun-types.js.map +1 -1
- package/dist/src/evals/rerun.js +30 -74
- package/dist/src/evals/rerun.js.map +1 -1
- package/dist/src/evals/result-helpers.js +13 -0
- package/dist/src/evals/result-helpers.js.map +1 -1
- package/dist/src/evals/retry-backoff.js +20 -0
- package/dist/src/evals/retry-backoff.js.map +1 -0
- package/dist/src/evals/retry-state.js +195 -0
- package/dist/src/evals/retry-state.js.map +1 -0
- package/dist/src/evals/scheduler-metrics.js +95 -0
- package/dist/src/evals/scheduler-metrics.js.map +1 -0
- package/dist/src/evals/scheduler-trace-replay.js +107 -0
- package/dist/src/evals/scheduler-trace-replay.js.map +1 -0
- package/dist/src/evals/service.js +87 -138
- package/dist/src/evals/service.js.map +1 -1
- package/dist/src/evals/trial-import.js +47 -50
- package/dist/src/evals/trial-import.js.map +1 -1
- package/dist/src/evals/trial-publication-recovery.js +45 -5
- package/dist/src/evals/trial-publication-recovery.js.map +1 -1
- package/dist/src/evals/trial-reference-validation.js +49 -0
- package/dist/src/evals/trial-reference-validation.js.map +1 -0
- package/dist/src/evals/verifier-artifacts.js +317 -0
- package/dist/src/evals/verifier-artifacts.js.map +1 -0
- package/dist/src/evals/verifier-eligibility.js +25 -0
- package/dist/src/evals/verifier-eligibility.js.map +1 -0
- package/dist/src/evals/verifier-only-rerun.js +210 -0
- package/dist/src/evals/verifier-only-rerun.js.map +1 -0
- package/dist/src/evals/verifier-runtime.js +61 -0
- package/dist/src/evals/verifier-runtime.js.map +1 -0
- package/dist/src/evals/verifier-score-artifacts.js +119 -0
- package/dist/src/evals/verifier-score-artifacts.js.map +1 -0
- package/dist/src/foundation/contained-file.js +85 -0
- package/dist/src/foundation/contained-file.js.map +1 -0
- package/dist/src/foundation/credential-redaction.js +26 -1
- package/dist/src/foundation/credential-redaction.js.map +1 -1
- package/dist/src/foundation/fs.js +2 -1
- package/dist/src/foundation/fs.js.map +1 -1
- package/dist/src/foundation/index.js +2 -1
- package/dist/src/foundation/index.js.map +1 -1
- package/dist/src/runs/adapter-process.js +12 -0
- package/dist/src/runs/adapter-process.js.map +1 -0
- package/dist/src/runs/executor.js +24 -30
- package/dist/src/runs/executor.js.map +1 -1
- package/dist/src/runs/finalizer.js +17 -0
- package/dist/src/runs/finalizer.js.map +1 -1
- package/dist/src/runs/index.js +5 -0
- package/dist/src/runs/index.js.map +1 -1
- package/dist/src/runs/phase-bundle.js +86 -0
- package/dist/src/runs/phase-bundle.js.map +1 -0
- package/dist/src/runs/phase-cancellation.js +107 -0
- package/dist/src/runs/phase-cancellation.js.map +1 -0
- package/dist/src/runs/phase-group.js +113 -0
- package/dist/src/runs/phase-group.js.map +1 -0
- package/dist/src/runs/query.js +1 -1
- package/dist/src/runs/query.js.map +1 -1
- package/dist/src/runs/records.js +3 -0
- package/dist/src/runs/records.js.map +1 -1
- package/dist/src/runs/request.js +4 -2
- package/dist/src/runs/request.js.map +1 -1
- package/dist/src/runs/verifier-evidence-redaction.js +61 -0
- package/dist/src/runs/verifier-evidence-redaction.js.map +1 -0
- package/dist/src/runs/verifier-evidence.js +459 -0
- package/dist/src/runs/verifier-evidence.js.map +1 -0
- package/dist/src/runs/verifier-structured-evidence.js +83 -0
- package/dist/src/runs/verifier-structured-evidence.js.map +1 -0
- package/dist/src/trajectories/analysis.js +288 -0
- package/dist/src/trajectories/analysis.js.map +1 -0
- package/dist/src/trajectories/chunk-projection.js +148 -0
- package/dist/src/trajectories/chunk-projection.js.map +1 -0
- package/dist/src/trajectories/content-projection.js +389 -0
- package/dist/src/trajectories/content-projection.js.map +1 -0
- package/dist/src/trajectories/dsh-chunk-contract.js +176 -0
- package/dist/src/trajectories/dsh-chunk-contract.js.map +1 -0
- package/dist/src/trajectories/dsh-contract.js +284 -0
- package/dist/src/trajectories/dsh-contract.js.map +1 -0
- package/dist/src/trajectories/events-chunk-drill.js +95 -0
- package/dist/src/trajectories/events-chunk-drill.js.map +1 -0
- package/dist/src/trajectories/events-page.js +367 -0
- package/dist/src/trajectories/events-page.js.map +1 -0
- package/dist/src/trajectories/index.js +3 -0
- package/dist/src/trajectories/index.js.map +1 -1
- package/dist/src/trajectories/request-attempt.js +47 -0
- package/dist/src/trajectories/request-attempt.js.map +1 -0
- package/dist/src/trajectories/stream-reader.js +311 -0
- package/dist/src/trajectories/stream-reader.js.map +1 -0
- package/dist/src/trajectories/surface-fold.js +126 -0
- package/dist/src/trajectories/surface-fold.js.map +1 -0
- package/dist/src/workers/remote-harbor-worker.js +3 -0
- package/dist/src/workers/remote-harbor-worker.js.map +1 -1
- package/docs/schemas/benchmark-hook-request-v1.schema.json +65 -0
- package/docs/schemas/benchmark-hook-response-v1.schema.json +75 -0
- package/docs/schemas/benchmark-hook-v1.schema.json +33 -0
- package/docs/schemas/benchmark-lock-v1.schema.json +351 -0
- package/docs/schemas/benchmark-metric-v1.schema.json +37 -0
- package/docs/schemas/benchmark-package-v1.schema.json +167 -0
- package/docs/schemas/benchmark-phase-group.schema.json +41 -0
- package/docs/schemas/benchmark-profile-v1.schema.json +141 -0
- package/docs/schemas/benchmark-task-v1.schema.json +384 -0
- package/docs/schemas/benchmark-tool-result-v1.schema.json +30 -0
- package/docs/schemas/error.schema.json +22 -0
- package/docs/schemas/eval-progress.schema.json +1 -15
- package/docs/schemas/eval-rerun-result.schema.json +6 -1
- package/docs/schemas/eval-rerun-submission.schema.json +3 -1
- package/docs/schemas/eval-result.schema.json +1 -15
- package/docs/schemas/eval-trial-publication.schema.json +1 -25
- package/docs/schemas/eval-trial-reference.schema.json +129 -0
- package/docs/schemas/regrade-assessment-reference.schema.json +11 -0
- package/docs/schemas/run-context.schema.json +15 -0
- package/docs/schemas/run-manifest.schema.json +4 -0
- package/docs/schemas/run-request.schema.json +4 -0
- package/docs/schemas/trajectory-analysis.schema.json +258 -0
- package/docs/schemas/trajectory-events-page.schema.json +40 -0
- package/docs/schemas/verifier-assessment.schema.json +77 -0
- package/docs/schemas/verifier-evidence.schema.json +276 -0
- package/integrations/harbor/hitch_benchmark.py +252 -0
- package/integrations/harbor/hitch_candidate_recycle.py +262 -0
- package/integrations/harbor/hitch_harbor_agent.py +666 -68
- package/integrations/harbor/hitch_harbor_environment.py +138 -0
- package/integrations/harbor/hitch_harbor_verifier.py +60 -2
- package/integrations/harbor/hitch_phase_supervisor.py +452 -0
- package/integrations/harbor/hitch_tool_client.mjs +91 -0
- package/integrations/model-call/cli.js +57 -0
- package/package.json +17 -6
|
@@ -0,0 +1,174 @@
|
|
|
1
|
+
import path from "node:path";
|
|
2
|
+
import { runInfrastructureRetries } from "./infrastructure-retry.js";
|
|
3
|
+
import { runRemoteInfrastructureRetries } from "./remote-infrastructure-retry.js";
|
|
4
|
+
import { beginPlannedInfrastructureRetry } from "./planned-retry-lifecycle.js";
|
|
5
|
+
import { localSourceBackendFailure } from "./result-helpers.js";
|
|
6
|
+
import { resourceRequirementForTask } from "./execution-plan-resources.js";
|
|
7
|
+
import { resolvedImageMapping } from "./environment-image-planning.js";
|
|
8
|
+
import { preparedArtifactForWorkItem } from "./work-item-artifacts.js";
|
|
9
|
+
import { workSchedulingPriority } from "./planned-execution-support.js";
|
|
10
|
+
import { ensurePhysicalRetryDecision, ensureTerminalRetryDecision, readEvalRetryState } from "./retry-state.js";
|
|
11
|
+
import { classifyTrialFailure, physicalRetryAllowed } from "./failure-classifier.js";
|
|
12
|
+
import { physicalRetryWorkItem } from "./physical-retry-work.js";
|
|
13
|
+
import { retryBackoffMs } from "./retry-backoff.js";
|
|
14
|
+
export async function runPlannedInfrastructureRetries(options, publisher, semaphore, metrics, workItem, completed) {
|
|
15
|
+
if (options.signal?.aborted || completed.run.backend.process_exit_code !== 0 || completed.run.rawResult === null)
|
|
16
|
+
return [];
|
|
17
|
+
if (options.localTransport && localSourceBackendFailure(completed.run.rawResult))
|
|
18
|
+
return [];
|
|
19
|
+
for (const trigger of completed.refs.filter((ref) => ref.observation_status === "invalid")) {
|
|
20
|
+
const classification = classifyTrialFailure(trigger);
|
|
21
|
+
if (physicalRetryAllowed(classification) && options.request.infrastructure_retries > 0)
|
|
22
|
+
continue;
|
|
23
|
+
const decision = await ensureTerminalRetryDecision({
|
|
24
|
+
evalDirectory: options.evalDirectory, evalId: options.evalId, item: workItem, retryIndex: 1, trigger,
|
|
25
|
+
exhausted: physicalRetryAllowed(classification),
|
|
26
|
+
});
|
|
27
|
+
if (decision)
|
|
28
|
+
options.sink.emit({
|
|
29
|
+
type: "eval.retry.skipped", work_id: workItem.work_id, decision_id: decision.decision_id,
|
|
30
|
+
task_id: trigger.task_id, attempt: trigger.attempt, disposition: decision.disposition,
|
|
31
|
+
classification: decision.classification, state: decision.state,
|
|
32
|
+
});
|
|
33
|
+
if (decision && providerCircuitFailure(decision.classification.code))
|
|
34
|
+
options.sink.emit({
|
|
35
|
+
type: "eval.provider-circuit.opened",
|
|
36
|
+
scope: "trial-retry",
|
|
37
|
+
mode: "retry-only",
|
|
38
|
+
provider: workItem.provider,
|
|
39
|
+
model: options.request.model || null,
|
|
40
|
+
work_id: workItem.work_id,
|
|
41
|
+
decision_id: decision.decision_id,
|
|
42
|
+
stable_code: decision.classification.code,
|
|
43
|
+
automatic_probe: false,
|
|
44
|
+
});
|
|
45
|
+
}
|
|
46
|
+
return runPlannedInfrastructureRetriesFromSource(options, publisher, semaphore, metrics, workItem, completed, 1);
|
|
47
|
+
}
|
|
48
|
+
function providerCircuitFailure(code) {
|
|
49
|
+
return code === "provider_quota_exhausted"
|
|
50
|
+
|| code === "provider_auth_failed"
|
|
51
|
+
|| code === "provider_configuration_invalid";
|
|
52
|
+
}
|
|
53
|
+
export async function runPlannedInfrastructureRetriesFromSource(options, publisher, semaphore, metrics, workItem, source, firstRetryIndex) {
|
|
54
|
+
const progress = publisher.current();
|
|
55
|
+
if (options.remoteWorkExecutor) {
|
|
56
|
+
return (await runRemoteInfrastructureRetries({
|
|
57
|
+
options, item: workItem, initial: source, progress, firstRetryIndex, metrics,
|
|
58
|
+
replaceProgressTrial: (ref, retryWorkId) => publisher.replaceInvalid(ref, retryWorkId),
|
|
59
|
+
currentProgress: () => publisher.current(),
|
|
60
|
+
})).runs;
|
|
61
|
+
}
|
|
62
|
+
const retries = await runInfrastructureRetries({
|
|
63
|
+
evalId: options.evalId,
|
|
64
|
+
evalDirectory: options.evalDirectory,
|
|
65
|
+
backendBaseDirectory: path.join(options.evalDirectory, "harbor", "work-items", source.workId, "infrastructure-retries"),
|
|
66
|
+
logicalAttempt: source.attempt,
|
|
67
|
+
initialRefs: source.refs,
|
|
68
|
+
progress,
|
|
69
|
+
request: options.request,
|
|
70
|
+
root: options.root,
|
|
71
|
+
resolvedRevision: options.resolvedRevision,
|
|
72
|
+
controllerRuntime: options.controllerRuntime,
|
|
73
|
+
preparedArtifact: preparedArtifactForWorkItem(options, workItem),
|
|
74
|
+
executionResources: resourceRequirementForTask(options.plan, source.tasks[0])?.main_limits ?? options.plan.default_trial_resources,
|
|
75
|
+
resolvedImages: resolvedImageMapping(workItem.image_refs ?? []),
|
|
76
|
+
...(source.environmentImages ? { environmentImages: source.environmentImages } : {}),
|
|
77
|
+
beginRetry: async ({ retry, triggers, backendDirectory }) => {
|
|
78
|
+
const directRelease = await semaphore?.acquire(options.signal, workSchedulingPriority(workItem));
|
|
79
|
+
try {
|
|
80
|
+
const lifecycle = await beginPlannedInfrastructureRetry({
|
|
81
|
+
options, item: workItem, retry, triggers, backendDirectory,
|
|
82
|
+
...(source.environmentImages ? { environmentImages: source.environmentImages } : {}),
|
|
83
|
+
});
|
|
84
|
+
return {
|
|
85
|
+
...lifecycle,
|
|
86
|
+
close: async () => { try {
|
|
87
|
+
await lifecycle.close();
|
|
88
|
+
}
|
|
89
|
+
finally {
|
|
90
|
+
directRelease?.();
|
|
91
|
+
} },
|
|
92
|
+
};
|
|
93
|
+
}
|
|
94
|
+
catch (error) {
|
|
95
|
+
directRelease?.();
|
|
96
|
+
throw error;
|
|
97
|
+
}
|
|
98
|
+
},
|
|
99
|
+
replaceProgressTrial: (ref, retryWorkId) => publisher.replaceInvalid(ref, retryWorkId),
|
|
100
|
+
currentProgress: () => publisher.current(),
|
|
101
|
+
originWorkItem: workItem,
|
|
102
|
+
firstRetryIndex,
|
|
103
|
+
onRetryExecutionStarted: (workId) => metrics.startWork(workId, "retry"),
|
|
104
|
+
onRetryExecutionFinished: (workId) => metrics.finishWork(workId),
|
|
105
|
+
onRetryBackoff: (durationMs) => metrics.addBlocked("backoff", durationMs),
|
|
106
|
+
onVerifierDuration: (durationMs) => metrics.addVerifier(durationMs),
|
|
107
|
+
...(options.onWorkItemQueued ? { onRetryWorkPlanned: options.onWorkItemQueued } : {}),
|
|
108
|
+
...(options.localTransport ? { localTransport: options.localTransport } : {}),
|
|
109
|
+
...(options.plan.model_capture ? { modelCapturePlan: options.plan.model_capture } : {}),
|
|
110
|
+
...(options.interactionCaptureExporter ? { interactionCaptureExporter: options.interactionCaptureExporter } : {}),
|
|
111
|
+
env: options.env,
|
|
112
|
+
...(options.harborExecutable !== undefined ? { harborExecutable: options.harborExecutable } : {}),
|
|
113
|
+
...(options.signal ? { signal: options.signal } : {}),
|
|
114
|
+
...(options.trialBundleGraceMs === undefined ? {} : { trialBundleGraceMs: options.trialBundleGraceMs }),
|
|
115
|
+
sink: options.sink,
|
|
116
|
+
...(options.localTransport ? { stopAfterResult: localSourceBackendFailure } : {}),
|
|
117
|
+
});
|
|
118
|
+
return retries.runs;
|
|
119
|
+
}
|
|
120
|
+
export async function reconcilePersistedInitialRetryDecisions(options) {
|
|
121
|
+
let state = await readEvalRetryState(options.evalDirectory, options.evalId);
|
|
122
|
+
for (const trigger of options.progress.trials.filter((trial) => trial.observation_status === "invalid")) {
|
|
123
|
+
if (state?.decisions.some((decision) => decision.trigger_trial_id === trigger.trial_id))
|
|
124
|
+
continue;
|
|
125
|
+
const item = options.plan.work_items.find((entry) => entry.logical_attempt === trigger.attempt && entry.task_ids.includes(trigger.task_id));
|
|
126
|
+
if (!item)
|
|
127
|
+
throw new TypeError(`persisted invalid trial is absent from execution plan: ${trigger.task_id}#${trigger.attempt}`);
|
|
128
|
+
const classification = classifyTrialFailure(trigger);
|
|
129
|
+
if (physicalRetryAllowed(classification) && options.request.infrastructure_retries > 0) {
|
|
130
|
+
const retryWork = physicalRetryWorkItem(item, 1, [trigger]);
|
|
131
|
+
await ensurePhysicalRetryDecision({
|
|
132
|
+
evalDirectory: options.evalDirectory, evalId: options.evalId, item, retryIndex: 1, trigger,
|
|
133
|
+
notBefore: new Date(Date.now() + retryBackoffMs(options.request.infrastructure_retry_backoff_ms, 1, retryWork.work_id)).toISOString(),
|
|
134
|
+
});
|
|
135
|
+
}
|
|
136
|
+
else {
|
|
137
|
+
await ensureTerminalRetryDecision({
|
|
138
|
+
evalDirectory: options.evalDirectory, evalId: options.evalId, item, retryIndex: 1, trigger,
|
|
139
|
+
exhausted: physicalRetryAllowed(classification),
|
|
140
|
+
});
|
|
141
|
+
}
|
|
142
|
+
state = await readEvalRetryState(options.evalDirectory, options.evalId);
|
|
143
|
+
}
|
|
144
|
+
}
|
|
145
|
+
export async function pendingRetryWork(options) {
|
|
146
|
+
const state = await readEvalRetryState(options.evalDirectory, options.evalId);
|
|
147
|
+
const pending = new Map();
|
|
148
|
+
if (!state)
|
|
149
|
+
return pending;
|
|
150
|
+
const seen = new Set();
|
|
151
|
+
for (const decision of state.decisions.filter((entry) => entry.disposition === "physical-retry" && entry.state === "planned")) {
|
|
152
|
+
if (!decision.retry_work_id || seen.has(decision.retry_work_id))
|
|
153
|
+
continue;
|
|
154
|
+
seen.add(decision.retry_work_id);
|
|
155
|
+
const item = options.plan.work_items.find((entry) => entry.slots.includes(decision.slot_id));
|
|
156
|
+
if (!item || item.logical_attempt === null || item.task_ids.length !== 1)
|
|
157
|
+
throw new TypeError(`retry decision slot is absent from execution plan: ${decision.slot_id}`);
|
|
158
|
+
const persistedTrigger = options.progress.trials.find((trial) => trial.trial_id === decision.trigger_trial_id);
|
|
159
|
+
const trigger = persistedTrigger ?? (decision.trigger_run_id ? {
|
|
160
|
+
trial_id: decision.trigger_trial_id, run_id: decision.trigger_run_id, task_id: item.task_ids[0],
|
|
161
|
+
attempt: item.logical_attempt, observation_status: "invalid", invalid_reason: decision.classification.code,
|
|
162
|
+
} : undefined);
|
|
163
|
+
if (!trigger)
|
|
164
|
+
throw new TypeError(`retry decision trigger cannot be recovered: ${decision.decision_id}`);
|
|
165
|
+
const taskId = item.task_ids[0];
|
|
166
|
+
const entries = pending.get(taskId) ?? [];
|
|
167
|
+
entries.push({ item, decision, trigger });
|
|
168
|
+
pending.set(taskId, entries);
|
|
169
|
+
}
|
|
170
|
+
for (const entries of pending.values())
|
|
171
|
+
entries.sort((left, right) => left.decision.retry_index - right.decision.retry_index);
|
|
172
|
+
return pending;
|
|
173
|
+
}
|
|
174
|
+
//# sourceMappingURL=planned-retry-execution.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"planned-retry-execution.js","sourceRoot":"","sources":["../../../src/evals/planned-retry-execution.ts"],"names":[],"mappings":"AAAA,OAAO,IAAI,MAAM,WAAW,CAAC;AAE7B,OAAO,EAAE,wBAAwB,EAAE,MAAM,2BAA2B,CAAC;AAErE,OAAO,EAAE,8BAA8B,EAAE,MAAM,kCAAkC,CAAC;AAClF,OAAO,EAAE,+BAA+B,EAAE,MAAM,8BAA8B,CAAC;AAC/E,OAAO,EAAE,yBAAyB,EAAE,MAAM,qBAAqB,CAAC;AAChE,OAAO,EAAE,0BAA0B,EAAE,MAAM,+BAA+B,CAAC;AAC3E,OAAO,EAAE,oBAAoB,EAAE,MAAM,iCAAiC,CAAC;AACvE,OAAO,EAAE,2BAA2B,EAAE,MAAM,0BAA0B,CAAC;AAEvE,OAAO,EAAE,sBAAsB,EAA0B,MAAM,gCAAgC,CAAC;AAKhG,OAAO,EAAE,2BAA2B,EAAE,2BAA2B,EAAE,kBAAkB,EAAE,MAAM,kBAAkB,CAAC;AAEhH,OAAO,EAAE,oBAAoB,EAAE,oBAAoB,EAAE,MAAM,yBAAyB,CAAC;AACrF,OAAO,EAAE,qBAAqB,EAAE,MAAM,0BAA0B,CAAC;AACjE,OAAO,EAAE,cAAc,EAAE,MAAM,oBAAoB,CAAC;AAQpD,MAAM,CAAC,KAAK,UAAU,+BAA+B,CACnD,OAAoC,EACpC,SAA4B,EAC5B,SAAwC,EACxC,OAA6B,EAC7B,QAA2B,EAC3B,SAA4B;IAE5B,IAAI,OAAO,CAAC,MAAM,EAAE,OAAO,IAAI,SAAS,CAAC,GAAG,CAAC,OAAO,CAAC,iBAAiB,KAAK,CAAC,IAAI,SAAS,CAAC,GAAG,CAAC,SAAS,KAAK,IAAI;QAAE,OAAO,EAAE,CAAC;IAC5H,IAAI,OAAO,CAAC,cAAc,IAAI,yBAAyB,CAAC,SAAS,CAAC,GAAG,CAAC,SAAS,CAAC;QAAE,OAAO,EAAE,CAAC;IAC5F,KAAK,MAAM,OAAO,IAAI,SAAS,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,kBAAkB,KAAK,SAAS,CAAC,EAAE,CAAC;QAC3F,MAAM,cAAc,GAAG,oBAAoB,CAAC,OAAO,CAAC,CAAC;QACrD,IAAI,oBAAoB,CAAC,cAAc,CAAC,IAAI,OAAO,CAAC,OAAO,CAAC,sBAAsB,GAAG,CAAC;YAAE,SAAS;QACjG,MAAM,QAAQ,GAAG,MAAM,2BAA2B,CAAC;YACjD,aAAa,EAAE,OAAO,CAAC,aAAa,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,IAAI,EAAE,QAAQ,EAAE,UAAU,EAAE,CAAC,EAAE,OAAO;YACpG,SAAS,EAAE,oBAAoB,CAAC,cAAc,CAAC;SAChD,CAAC,CAAC;QACH,IAAI,QAAQ;YAAE,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC;gBAC9B,IAAI,EAAE,oBAAoB,EAAE,OAAO,EAAE,QAAQ,CAAC,OAAO,EAAE,WAAW,EAAE,QAAQ,CAAC,WAAW;gBACxF,OAAO,EAAE,OAAO,CAAC,OAAO,EAAE,OAAO,EAAE,OAAO,CAAC,OAAO,EAAE,WAAW,EAAE,QAAQ,CAAC,WAAW;gBACrF,cAAc,EAAE,QAAQ,CAAC,cAAc,EAAE,KAAK,EAAE,QAAQ,CAAC,KAAK;aAC/D,CAAC,CAAC;QACH,IAAI,QAAQ,IAAI,sBAAsB,CAAC,QAAQ,CAAC,cAAc,CAAC,IAAI,CAAC;YAAE,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC;gBACtF,IAAI,EAAE,8BAA8B;gBACpC,KAAK,EAAE,aAAa;gBACpB,IAAI,EAAE,YAAY;gBAClB,QAAQ,EAAE,QAAQ,CAAC,QAAQ;gBAC3B,KAAK,EAAE,OAAO,CAAC,OAAO,CAAC,KAAK,IAAI,IAAI;gBACpC,OAAO,EAAE,QAAQ,CAAC,OAAO;gBACzB,WAAW,EAAE,QAAQ,CAAC,WAAW;gBACjC,WAAW,EAAE,QAAQ,CAAC,cAAc,CAAC,IAAI;gBACzC,eAAe,EAAE,KAAK;aACvB,CAAC,CAAC;IACL,CAAC;IACD,OAAO,yCAAyC,CAAC,OAAO,EAAE,SAAS,EAAE,SAAS,EAAE,OAAO,EAAE,QAAQ,EAAE,SAAS,EAAE,CAAC,CAAC,CAAC;AACnH,CAAC;AAED,SAAS,sBAAsB,CAAC,IAAY;IAC1C,OAAO,IAAI,KAAK,0BAA0B;WACrC,IAAI,KAAK,sBAAsB;WAC/B,IAAI,KAAK,gCAAgC,CAAC;AACjD,CAAC;AAUD,MAAM,CAAC,KAAK,UAAU,yCAAyC,CAC7D,OAAoC,EACpC,SAA4B,EAC5B,SAAwC,EACxC,OAA6B,EAC7B,QAA2B,EAC3B,MAAmB,EACnB,eAAuB;IAEvB,MAAM,QAAQ,GAAG,SAAS,CAAC,OAAO,EAAE,CAAC;IACrC,IAAI,OAAO,CAAC,kBAAkB,EAAE,CAAC;QAC/B,OAAO,CAAC,MAAM,8BAA8B,CAAC;YAC3C,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,QAAQ,EAAE,eAAe,EAAE,OAAO;YAC5E,oBAAoB,EAAE,CAAC,GAAG,EAAE,WAAW,EAAE,EAAE,CAAC,SAAS,CAAC,cAAc,CAAC,GAAG,EAAE,WAAW,CAAC;YACtF,eAAe,EAAE,GAAG,EAAE,CAAC,SAAS,CAAC,OAAO,EAAE;SAC3C,CAAC,CAAC,CAAC,IAAI,CAAC;IACX,CAAC;IACD,MAAM,OAAO,GAAG,MAAM,wBAAwB,CAAC;QAC7C,MAAM,EAAE,OAAO,CAAC,MAAM;QACtB,aAAa,EAAE,OAAO,CAAC,aAAa;QACpC,oBAAoB,EAAE,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,aAAa,EAAE,QAAQ,EAAE,YAAY,EAAE,MAAM,CAAC,MAAM,EAAE,wBAAwB,CAAC;QACvH,cAAc,EAAE,MAAM,CAAC,OAAO;QAC9B,WAAW,EAAE,MAAM,CAAC,IAAI;QACxB,QAAQ;QACR,OAAO,EAAE,OAAO,CAAC,OAAO;QACxB,IAAI,EAAE,OAAO,CAAC,IAAI;QAClB,gBAAgB,EAAE,OAAO,CAAC,gBAAgB;QAC1C,iBAAiB,EAAE,OAAO,CAAC,iBAAiB;QAC5C,gBAAgB,EAAE,2BAA2B,CAAC,OAAO,EAAE,QAAQ,CAAC;QAChE,kBAAkB,EAAE,0BAA0B,CAAC,OAAO,CAAC,IAAI,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC,CAAW,CAAC,EAAE,WAAW,IAAI,OAAO,CAAC,IAAI,CAAC,uBAAuB;QAC5I,cAAc,EAAE,oBAAoB,CAAC,QAAQ,CAAC,UAAU,IAAI,EAAE,CAAC;QAC/D,GAAG,CAAC,MAAM,CAAC,iBAAiB,CAAC,CAAC,CAAC,EAAE,iBAAiB,EAAE,MAAM,CAAC,iBAAiB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACpF,UAAU,EAAE,KAAK,EAAE,EAAE,KAAK,EAAE,QAAQ,EAAE,gBAAgB,EAAE,EAAE,EAAE;YAC1D,MAAM,aAAa,GAAG,MAAM,SAAS,EAAE,OAAO,CAAC,OAAO,CAAC,MAAM,EAAE,sBAAsB,CAAC,QAAQ,CAAC,CAAC,CAAC;YACjG,IAAI,CAAC;gBACH,MAAM,SAAS,GAAG,MAAM,+BAA+B,CAAC;oBACtD,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,KAAK,EAAE,QAAQ,EAAE,gBAAgB;oBAC1D,GAAG,CAAC,MAAM,CAAC,iBAAiB,CAAC,CAAC,CAAC,EAAE,iBAAiB,EAAE,MAAM,CAAC,iBAAiB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;iBACrF,CAAC,CAAC;gBACH,OAAO;oBACL,GAAG,SAAS;oBACZ,KAAK,EAAE,KAAK,IAAI,EAAE,GAAG,IAAI,CAAC;wBAAC,MAAM,SAAS,CAAC,KAAK,EAAE,CAAC;oBAAC,CAAC;4BAAS,CAAC;wBAAC,aAAa,EAAE,EAAE,CAAC;oBAAC,CAAC,CAAC,CAAC;iBACvF,CAAC;YACJ,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,aAAa,EAAE,EAAE,CAAC;gBAClB,MAAM,KAAK,CAAC;YACd,CAAC;QACH,CAAC;QACD,oBAAoB,EAAE,CAAC,GAAG,EAAE,WAAW,EAAE,EAAE,CAAC,SAAS,CAAC,cAAc,CAAC,GAAG,EAAE,WAAW,CAAC;QACtF,eAAe,EAAE,GAAG,EAAE,CAAC,SAAS,CAAC,OAAO,EAAE;QAC1C,cAAc,EAAE,QAAQ;QACxB,eAAe;QACf,uBAAuB,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC,OAAO,CAAC,SAAS,CAAC,MAAM,EAAE,OAAO,CAAC;QACvE,wBAAwB,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC,OAAO,CAAC,UAAU,CAAC,MAAM,CAAC;QAChE,cAAc,EAAE,CAAC,UAAU,EAAE,EAAE,CAAC,OAAO,CAAC,UAAU,CAAC,SAAS,EAAE,UAAU,CAAC;QACzE,kBAAkB,EAAE,CAAC,UAAU,EAAE,EAAE,CAAC,OAAO,CAAC,WAAW,CAAC,UAAU,CAAC;QACnE,GAAG,CAAC,OAAO,CAAC,gBAAgB,CAAC,CAAC,CAAC,EAAE,kBAAkB,EAAE,OAAO,CAAC,gBAAgB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACrF,GAAG,CAAC,OAAO,CAAC,cAAc,CAAC,CAAC,CAAC,EAAE,cAAc,EAAE,OAAO,CAAC,cAAc,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC7E,GAAG,CAAC,OAAO,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC,CAAC,EAAE,gBAAgB,EAAE,OAAO,CAAC,IAAI,CAAC,aAAa,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACvF,GAAG,CAAC,OAAO,CAAC,0BAA0B,CAAC,CAAC,CAAC,EAAE,0BAA0B,EAAE,OAAO,CAAC,0BAA0B,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACjH,GAAG,EAAE,OAAO,CAAC,GAAG;QAChB,GAAG,CAAC,OAAO,CAAC,gBAAgB,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,gBAAgB,EAAE,OAAO,CAAC,gBAAgB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACjG,GAAG,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACrD,GAAG,CAAC,OAAO,CAAC,kBAAkB,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,kBAAkB,EAAE,OAAO,CAAC,kBAAkB,EAAE,CAAC;QACvG,IAAI,EAAE,OAAO,CAAC,IAAI;QAClB,GAAG,CAAC,OAAO,CAAC,cAAc,CAAC,CAAC,CAAC,EAAE,eAAe,EAAE,yBAAyB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAClF,CAAC,CAAC;IACH,OAAO,OAAO,CAAC,IAAI,CAAC;AACtB,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,uCAAuC,CAAC,OAAoC;IAChG,IAAI,KAAK,GAAG,MAAM,kBAAkB,CAAC,OAAO,CAAC,aAAa,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC;IAC5E,KAAK,MAAM,OAAO,IAAI,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,kBAAkB,KAAK,SAAS,CAAC,EAAE,CAAC;QACxG,IAAI,KAAK,EAAE,SAAS,CAAC,IAAI,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,gBAAgB,KAAK,OAAO,CAAC,QAAQ,CAAC;YAAE,SAAS;QAClG,MAAM,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,eAAe,KAAK,OAAO,CAAC,OAAO,IAAI,KAAK,CAAC,QAAQ,CAAC,QAAQ,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC;QAC5I,IAAI,CAAC,IAAI;YAAE,MAAM,IAAI,SAAS,CAAC,0DAA0D,OAAO,CAAC,OAAO,IAAI,OAAO,CAAC,OAAO,EAAE,CAAC,CAAC;QAC/H,MAAM,cAAc,GAAG,oBAAoB,CAAC,OAAO,CAAC,CAAC;QACrD,IAAI,oBAAoB,CAAC,cAAc,CAAC,IAAI,OAAO,CAAC,OAAO,CAAC,sBAAsB,GAAG,CAAC,EAAE,CAAC;YACvF,MAAM,SAAS,GAAG,qBAAqB,CAAC,IAAI,EAAE,CAAC,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC;YAC5D,MAAM,2BAA2B,CAAC;gBAChC,aAAa,EAAE,OAAO,CAAC,aAAa,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,IAAI,EAAE,UAAU,EAAE,CAAC,EAAE,OAAO;gBAC1F,SAAS,EAAE,IAAI,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE,GAAG,cAAc,CAAC,OAAO,CAAC,OAAO,CAAC,+BAA+B,EAAE,CAAC,EAAE,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC,WAAW,EAAE;aACtI,CAAC,CAAC;QACL,CAAC;aAAM,CAAC;YACN,MAAM,2BAA2B,CAAC;gBAChC,aAAa,EAAE,OAAO,CAAC,aAAa,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,IAAI,EAAE,UAAU,EAAE,CAAC,EAAE,OAAO;gBAC1F,SAAS,EAAE,oBAAoB,CAAC,cAAc,CAAC;aAChD,CAAC,CAAC;QACL,CAAC;QACD,KAAK,GAAG,MAAM,kBAAkB,CAAC,OAAO,CAAC,aAAa,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC;IAC1E,CAAC;AACH,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,gBAAgB,CAAC,OAAoC;IACzE,MAAM,KAAK,GAAG,MAAM,kBAAkB,CAAC,OAAO,CAAC,aAAa,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC;IAC9E,MAAM,OAAO,GAAG,IAAI,GAAG,EAA8B,CAAC;IACtD,IAAI,CAAC,KAAK;QAAE,OAAO,OAAO,CAAC;IAC3B,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAC;IAC/B,KAAK,MAAM,QAAQ,IAAI,KAAK,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,WAAW,KAAK,gBAAgB,IAAI,KAAK,CAAC,KAAK,KAAK,SAAS,CAAC,EAAE,CAAC;QAC9H,IAAI,CAAC,QAAQ,CAAC,aAAa,IAAI,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,aAAa,CAAC;YAAE,SAAS;QAC1E,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,aAAa,CAAC,CAAC;QACjC,MAAM,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,KAAK,CAAC,QAAQ,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC;QAC7F,IAAI,CAAC,IAAI,IAAI,IAAI,CAAC,eAAe,KAAK,IAAI,IAAI,IAAI,CAAC,QAAQ,CAAC,MAAM,KAAK,CAAC;YAAE,MAAM,IAAI,SAAS,CAAC,sDAAsD,QAAQ,CAAC,OAAO,EAAE,CAAC,CAAC;QACxK,MAAM,gBAAgB,GAAG,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,QAAQ,KAAK,QAAQ,CAAC,gBAAgB,CAAC,CAAC;QAC/G,MAAM,OAAO,GAAG,gBAAgB,IAAI,CAAC,QAAQ,CAAC,cAAc,CAAC,CAAC,CAAC;YAC7D,QAAQ,EAAE,QAAQ,CAAC,gBAAgB,EAAE,MAAM,EAAE,QAAQ,CAAC,cAAc,EAAE,OAAO,EAAE,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAW;YACzG,OAAO,EAAE,IAAI,CAAC,eAAe,EAAE,kBAAkB,EAAE,SAAkB,EAAE,cAAc,EAAE,QAAQ,CAAC,cAAc,CAAC,IAAI;SACpH,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC;QACf,IAAI,CAAC,OAAO;YAAE,MAAM,IAAI,SAAS,CAAC,+CAA+C,QAAQ,CAAC,WAAW,EAAE,CAAC,CAAC;QACzG,MAAM,MAAM,GAAG,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAW,CAAC;QAC1C,MAAM,OAAO,GAAG,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC;QAC1C,OAAO,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,CAAC;QAC1C,OAAO,CAAC,GAAG,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IAC/B,CAAC;IACD,KAAK,MAAM,OAAO,IAAI,OAAO,CAAC,MAAM,EAAE;QAAE,OAAO,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,QAAQ,CAAC,WAAW,GAAG,KAAK,CAAC,QAAQ,CAAC,WAAW,CAAC,CAAC;IAC9H,OAAO,OAAO,CAAC;AACjB,CAAC"}
|
|
@@ -5,8 +5,10 @@ import { resourceRequirementForTask, runtimeResourcesForTask } from "./execution
|
|
|
5
5
|
import { recordLocalDockerProcessExit, recordLocalDockerProcessStart, releaseLocalDockerProcessRecord } from "./local-docker-provider.js";
|
|
6
6
|
import { resolvedImageMapping } from "./environment-image-planning.js";
|
|
7
7
|
import { prebuiltTaskImage, verifyTrialEnvironmentImageExecution } from "./trial-environment-evidence.js";
|
|
8
|
+
import { physicalRetryWorkItem } from "./physical-retry-work.js";
|
|
8
9
|
export async function beginPlannedInfrastructureRetry(input) {
|
|
9
|
-
const { options
|
|
10
|
+
const { options } = input;
|
|
11
|
+
const item = physicalRetryWorkItem(input.item, input.retry, input.triggers);
|
|
10
12
|
const taskId = item.task_ids[0];
|
|
11
13
|
let permit;
|
|
12
14
|
let lease;
|
|
@@ -52,7 +54,7 @@ export async function beginPlannedInfrastructureRetry(input) {
|
|
|
52
54
|
attempt: item.logical_attempt,
|
|
53
55
|
trigger_trials: input.triggers.map((trial) => trial.trial_id).sort(),
|
|
54
56
|
});
|
|
55
|
-
return retryLifecycle(input, lease, permit, runningAnnounced);
|
|
57
|
+
return retryLifecycle({ ...input, item }, lease, permit, runningAnnounced);
|
|
56
58
|
}
|
|
57
59
|
catch (error) {
|
|
58
60
|
if (lease)
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"planned-retry-lifecycle.js","sourceRoot":"","sources":["../../../src/evals/planned-retry-lifecycle.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,uBAAuB,EAAE,MAAM,uBAAuB,CAAC;AAChE,OAAO,EAAE,2BAA2B,EAAE,MAAM,+BAA+B,CAAC;AAC5E,OAAO,EAAE,oCAAoC,EAAE,8BAA8B,EAAE,oBAAoB,EAAE,MAAM,uBAAuB,CAAC;AAEnI,OAAO,EAAE,0BAA0B,EAAE,uBAAuB,EAAE,MAAM,+BAA+B,CAAC;AAEpG,OAAO,EAAE,4BAA4B,EAAE,6BAA6B,EAAE,+BAA+B,EAAE,MAAM,4BAA4B,CAAC;AAE1I,OAAO,EAAE,oBAAoB,EAAE,MAAM,iCAAiC,CAAC;AACvE,OAAO,EAAE,iBAAiB,EAAE,oCAAoC,EAAE,MAAM,iCAAiC,CAAC;
|
|
1
|
+
{"version":3,"file":"planned-retry-lifecycle.js","sourceRoot":"","sources":["../../../src/evals/planned-retry-lifecycle.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,uBAAuB,EAAE,MAAM,uBAAuB,CAAC;AAChE,OAAO,EAAE,2BAA2B,EAAE,MAAM,+BAA+B,CAAC;AAC5E,OAAO,EAAE,oCAAoC,EAAE,8BAA8B,EAAE,oBAAoB,EAAE,MAAM,uBAAuB,CAAC;AAEnI,OAAO,EAAE,0BAA0B,EAAE,uBAAuB,EAAE,MAAM,+BAA+B,CAAC;AAEpG,OAAO,EAAE,4BAA4B,EAAE,6BAA6B,EAAE,+BAA+B,EAAE,MAAM,4BAA4B,CAAC;AAE1I,OAAO,EAAE,oBAAoB,EAAE,MAAM,iCAAiC,CAAC;AACvE,OAAO,EAAE,iBAAiB,EAAE,oCAAoC,EAAE,MAAM,iCAAiC,CAAC;AAE1G,OAAO,EAAE,qBAAqB,EAAE,MAAM,0BAA0B,CAAC;AAEjE,MAAM,CAAC,KAAK,UAAU,+BAA+B,CAAC,KAOrD;IACC,MAAM,EAAE,OAAO,EAAE,GAAG,KAAK,CAAC;IAC1B,MAAM,IAAI,GAAG,qBAAqB,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,KAAK,EAAE,KAAK,CAAC,QAAQ,CAAC,CAAC;IAC5E,MAAM,MAAM,GAAG,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAW,CAAC;IAC1C,IAAI,MAAyF,CAAC;IAC9F,IAAI,KAAuC,CAAC;IAC5C,IAAI,gBAAgB,GAAG,KAAK,CAAC;IAC7B,IAAI,CAAC;QACH,MAAM,GAAG,MAAM,OAAO,CAAC,SAAS,EAAE,OAAO,CAAC;YACxC,MAAM,EAAE,OAAO,CAAC,MAAM;YACtB,QAAQ,EAAE,IAAI;YACd,cAAc,EAAE,OAAO,CAAC,IAAI,CAAC,eAAe;YAC5C,GAAG,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SACtD,CAAC,CAAC;QACH,IAAI,MAAM;YAAE,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC;gBAC5B,IAAI,EAAE,8BAA8B;gBACpC,cAAc,EAAE,+BAA+B;gBAC/C,OAAO,EAAE,IAAI,CAAC,OAAO;gBACrB,KAAK,EAAE,KAAK,CAAC,KAAK;gBAClB,aAAa,EAAE,MAAM,CAAC,YAAY;gBAClC,cAAc,EAAE,MAAM,CAAC,aAAa;gBACpC,WAAW,EAAE,IAAI,CAAC,WAAW;aAC9B,CAAC,CAAC;QACH,KAAK,GAAG,MAAM,oBAAoB,CAAC;YACjC,aAAa,EAAE,OAAO,CAAC,aAAa;YACpC,MAAM,EAAE,OAAO,CAAC,MAAM;YACtB,MAAM,EAAE,IAAI,CAAC,OAAO;YACpB,MAAM,EAAE,MAAM,CAAC,CAAC,CAAC,EAAE,GAAG,OAAO,CAAC,MAAM,EAAE,kBAAkB,EAAE,MAAM,CAAC,YAAY,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,MAAM;YAChG,WAAW,EAAE,IAAI,CAAC,WAAW;YAC7B,KAAK,EAAE,8BAA8B;SACtC,CAAC,CAAC;QACH,MAAM,KAAK,GAAG,KAAK,CAAC,OAAO,EAAE,CAAC,KAAK,CAAC;QACpC,MAAM,KAAK,CAAC,WAAW,CAAC,KAAK,CAAC,CAAC;QAC/B,MAAM,OAAO,CAAC,eAAe,EAAE,CAAC,IAAI,CAAC,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC;QACxE,gBAAgB,GAAG,IAAI,CAAC;QACxB,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC;YAChB,IAAI,EAAE,6BAA6B;YACnC,cAAc,EAAE,+BAA+B;YAC/C,kBAAkB,EAAE,IAAI;YACxB,OAAO,EAAE,IAAI,CAAC,OAAO;YACrB,QAAQ,EAAE,KAAK,CAAC,OAAO;YACvB,WAAW,EAAE,KAAK;YAClB,KAAK,EAAE,KAAK,CAAC,KAAK;YAClB,OAAO,EAAE,MAAM;YACf,OAAO,EAAE,IAAI,CAAC,eAAe;YAC7B,cAAc,EAAE,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,IAAI,EAAE;SACrE,CAAC,CAAC;QACH,OAAO,cAAc,CAAC,EAAE,GAAG,KAAK,EAAE,IAAI,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,gBAAgB,CAAC,CAAC;IAC7E,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAI,KAAK;YAAE,MAAM,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,SAAS,CAAC,CAAC;QAC7E,IAAI,gBAAgB;YAAE,MAAM,OAAO,CAAC,eAAe,EAAE,CAAC,IAAI,CAAC,OAAO,EAAE,KAAK,EAAE,OAAO,IAAI,EAAE,EAAE,UAAU,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,SAAS,CAAC,CAAC;QAC7H,MAAM,EAAE,OAAO,EAAE,CAAC;QAClB,MAAM,KAAK,CAAC;IACd,CAAC;AACH,CAAC;AAED,SAAS,cAAc,CACrB,KAMC,EACD,KAA2B,EAC3B,MAAyG,EACzG,gBAAyB;IAEzB,MAAM,EAAE,OAAO,EAAE,IAAI,EAAE,GAAG,KAAK,CAAC;IAChC,MAAM,MAAM,GAAG,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAW,CAAC;IAC1C,MAAM,KAAK,GAAG,KAAK,CAAC,OAAO,EAAE,CAAC,KAAK,CAAC;IACpC,MAAM,OAAO,GAAG,uBAAuB,CAAC,OAAO,CAAC,IAAI,EAAE,MAAM,EAAE,IAAI,CAAC,WAAW,CAAC,CAAC;IAChF,MAAM,SAAS,GAAG,uBAAuB,CAAC,OAAO,CAAC,IAAI,EAAE,KAAK,CAAC,OAAO,EAAE,EAAE,MAAM,CAAC,CAAC;IACjF,MAAM,QAAQ,GAAG,2BAA2B,CAAC;QAC3C,SAAS;QACT,QAAQ,EAAE,OAAO,CAAC,MAAM,CAAC,QAAQ;QACjC,iBAAiB,EAAE,OAAO,CAAC,MAAM,CAAC,iBAAiB;QACnD,WAAW,EAAE,IAAI,CAAC,WAAW;QAC7B,UAAU,EAAE,OAAO,CAAC,UAAU;QAC9B,aAAa,EAAE,OAAO,CAAC,aAAa;QACpC,GAAG,EAAE,OAAO,CAAC,GAAG;QAChB,GAAG,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KACtD,CAAC,CAAC;IACH,IAAI,eAAe,GAAG,KAAK,CAAC;IAC5B,IAAI,gBAAyB,CAAC;IAC9B,IAAI,aAAa,GAAG,OAAO,CAAC,OAAO,EAAE,CAAC;IACtC,MAAM,KAAK,GAAG,WAAW,CAAC,GAAG,EAAE;QAC7B,IAAI,gBAAgB,KAAK,SAAS;YAAE,OAAO;QAC3C,aAAa,GAAG,aAAa,CAAC,IAAI,CAAC,KAAK,IAAI,EAAE;YAC5C,MAAM,OAAO,GAAG,MAAM,KAAK,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;YAC7C,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC;gBAChB,IAAI,EAAE,eAAe;gBACrB,cAAc,EAAE,+BAA+B;gBAC/C,OAAO,EAAE,IAAI,CAAC,OAAO;gBACrB,QAAQ,EAAE,KAAK,CAAC,OAAO;gBACvB,WAAW,EAAE,OAAO,CAAC,KAAK;gBAC1B,YAAY,EAAE,OAAO,CAAC,YAAY;gBAClC,UAAU,EAAE,OAAO,CAAC,UAAU;aAC/B,CAAC,CAAC;QACL,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,KAAK,EAAE,EAAE,GAAG,gBAAgB,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;IACvD,CAAC,EAAE,oCAAoC,CAAC,CAAC;IACzC,KAAK,CAAC,KAAK,EAAE,CAAC;IACd,IAAI,MAAiC,CAAC;IACtC,MAAM,aAAa,GAAG,iBAAiB,CAAC,KAAK,CAAC,iBAAiB,CAAC,CAAC;IACjE,MAAM,KAAK,GAAG,GAAkB,EAAE,CAAC,MAAM,KAAK,mBAAmB,CAAC,EAAE,KAAK,EAAE,KAAK,EAAE,MAAM,EAAE,QAAQ,EAAE,KAAK,EAAE,aAAa,EAAE,GAAG,EAAE,CAAC,aAAa,EAAE,gBAAgB,EAAE,GAAG,EAAE,CAAC,gBAAgB,EAAE,eAAe,EAAE,GAAG,EAAE,CAAC,eAAe,EAAE,gBAAgB,EAAE,CAAC,CAAC;IACrP,OAAO;QACL,OAAO,EAAE,KAAK,CAAC,OAAO;QACtB,MAAM,EAAE,IAAI,CAAC,OAAO;QACpB,GAAG,CAAC,KAAK,CAAC,iBAAiB,CAAC,CAAC,CAAC,EAAE,iBAAiB,EAAE,KAAK,CAAC,iBAAiB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAClF,wBAAwB,EAAE,KAAK,IAAI,EAAE;YACnC,IAAI,gBAAgB,KAAK,SAAS;gBAAE,MAAM,gBAAgB,CAAC;YAC3D,OAAO,oCAAoC,CAAC,MAAM,QAAQ,CAAC,OAAO,EAAE,EAAE,KAAK,CAAC,iBAAiB,CAAC,CAAC;QACjG,CAAC;QACD,OAAO,EAAE;YACP,kBAAkB,EAAE,0BAA0B,CAAC,OAAO,CAAC,IAAI,EAAE,MAAM,CAAC,EAAE,WAAW,IAAI,OAAO,CAAC,IAAI,CAAC,uBAAuB;YACzH,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,aAAa,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,mBAAmB,EAAE,OAAO,CAAC,aAAa,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACxG,eAAe,EAAE,SAAS;YAC1B,cAAc,EAAE,oBAAoB,CAAC,IAAI,CAAC,UAAU,IAAI,EAAE,CAAC;YAC3D,GAAG,CAAC,aAAa,CAAC,CAAC,CAAC,EAAE,iBAAiB,EAAE,aAAa,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YAC9D,GAAG,CAAC,OAAO,CAAC,MAAM,CAAC,QAAQ,KAAK,cAAc,IAAI,OAAO,CAAC,QAAQ,KAAK,OAAO,CAAC,CAAC,CAAC;gBAC/E,kBAAkB,EAAE,IAAI;gBACxB,gBAAgB,EAAE,CAAC,GAAW,EAAE,EAAE,CAAC,6BAA6B,CAAC;oBAC/D,IAAI,EAAE,OAAO,CAAC,IAAI;oBAClB,QAAQ,EAAE,OAAO,CAAC,MAAM,CAAC,QAAQ;oBACjC,aAAa,EAAE,OAAO,CAAC,aAAa;oBACpC,KAAK,EAAE,KAAK,CAAC,OAAO,EAAE;oBACtB,gBAAgB,EAAE,KAAK,CAAC,gBAAgB;oBACxC,GAAG;iBACJ,CAAC,CAAC,IAAI,CAAC,GAAG,EAAE,GAAG,eAAe,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC;gBAC1C,eAAe,EAAE,CAAC,MAA8D,EAAE,EAAE,CAAC,4BAA4B,CAAC;oBAChH,IAAI,EAAE,OAAO,CAAC,IAAI;oBAClB,OAAO,EAAE,KAAK,CAAC,OAAO;oBACtB,KAAK;oBACL,GAAG,MAAM;iBACV,CAAC,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC,SAAS,CAAC;aACzB,CAAC,CAAC,CAAC,EAAE,CAAC;SACR;QACD,KAAK;KACN,CAAC;AACJ,CAAC;AAED,KAAK,UAAU,mBAAmB,CAAC,KAUlC;IACC,MAAM,EAAE,OAAO,EAAE,IAAI,EAAE,GAAG,KAAK,CAAC,KAAK,CAAC;IACtC,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC,KAAK,CAAC;IAC1C,IAAI,OAAgB,CAAC;IACrB,MAAM,OAAO,GAAG,KAAK,EAAE,SAAiC,EAAiB,EAAE;QACzE,IAAI,CAAC;YAAC,MAAM,SAAS,EAAE,CAAC;QAAC,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAAC,OAAO,KAAK,KAAK,CAAC;QAAC,CAAC;IACjE,CAAC,CAAC;IACF,aAAa,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;IAC3B,MAAM,OAAO,CAAC,GAAG,EAAE,CAAC,KAAK,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC,CAAC;IAC3C,MAAM,OAAO,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC;IACnC,OAAO,KAAK,KAAK,CAAC,gBAAgB,EAAE,CAAC;IACrC,IAAI,KAAK,CAAC,eAAe,EAAE;QAAE,MAAM,OAAO,CAAC,GAAG,EAAE,CAAC,+BAA+B,CAAC,EAAE,IAAI,EAAE,OAAO,CAAC,IAAI,EAAE,OAAO,EAAE,KAAK,CAAC,KAAK,CAAC,OAAO,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC;IAC/I,MAAM,OAAO,CAAC,KAAK,IAAI,EAAE;QACvB,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;QAClD,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC;YAChB,IAAI,EAAE,oCAAoC;YAC1C,cAAc,EAAE,+BAA+B;YAC/C,OAAO,EAAE,IAAI,CAAC,OAAO;YACrB,QAAQ,EAAE,KAAK,CAAC,KAAK,CAAC,OAAO;YAC7B,WAAW,EAAE,QAAQ,CAAC,KAAK;YAC3B,KAAK,EAAE,KAAK,CAAC,KAAK,CAAC,KAAK;YACxB,KAAK,EAAE,QAAQ,CAAC,KAAK;SACtB,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IACH,IAAI,OAAO,CAAC,oBAAoB,EAAE,CAAC;QACjC,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,oBAAoB,CAAC,EAAE,IAAI,EAAE,OAAO,CAAC,IAAI,EAAE,QAAQ,EAAE,CAAC,KAAK,CAAC,KAAK,CAAC,OAAO,CAAC,EAAE,GAAG,EAAE,OAAO,CAAC,GAAG,EAAE,CAAC,CAAC;YAC7H,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,sCAAsC,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,QAAQ,EAAE,KAAK,CAAC,KAAK,CAAC,OAAO,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC;QAC7O,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,mCAAmC,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,QAAQ,EAAE,KAAK,CAAC,KAAK,CAAC,OAAO,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,CAAC,KAAK,EAAE,IAAI,EAAG,KAA2B,CAAC,IAAI,IAAI,sBAAsB,EAAE,CAAC,CAAC;QACtN,CAAC;IACH,CAAC;IACD,IAAI,KAAK,CAAC,gBAAgB;QAAE,MAAM,OAAO,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,eAAe,EAAE,CAAC,IAAI,CAAC,OAAO,EAAE,KAAK,CAAC,KAAK,CAAC,OAAO,EAAE,UAAU,CAAC,IAAI,OAAO,CAAC,OAAO,EAAE,CAAC,CAAC;IAC/I,KAAK,CAAC,MAAM,EAAE,OAAO,EAAE,CAAC;IACxB,IAAI,OAAO,KAAK,SAAS;QAAE,MAAM,OAAO,CAAC;AAC3C,CAAC"}
|
|
@@ -0,0 +1,225 @@
|
|
|
1
|
+
import { cp, lstat, readdir, rename } from "node:fs/promises";
|
|
2
|
+
import path from "node:path";
|
|
3
|
+
import { HitchError, SCHEMA_VERSION, atomicWriteJSON, ensureDir, readJSON } from "../foundation/index.js";
|
|
4
|
+
import { logicalPlanModelCapture, readEvalLogicalPlan } from "./eval-logical-plan.js";
|
|
5
|
+
import { readEvalProgress } from "./progress.js";
|
|
6
|
+
import { resolveLocalDatasetTaskIds, validateEvalId } from "./request.js";
|
|
7
|
+
import { evalRerunSemantics } from "./rerun-types.js";
|
|
8
|
+
import { invalidTrialSlots, slotKey, sortSlots, uniqueTasks, validateProgressPlan } from "./rerun-slots.js";
|
|
9
|
+
import { runEval } from "./service.js";
|
|
10
|
+
/** Restart an eval that failed before its executable plan was fully materialized. */
|
|
11
|
+
export async function restartIncompleteEval(options, request, startedAt, rerunDirectory, statePath) {
|
|
12
|
+
const materializedFiles = ["plan.json", "execution-plan.json", "progress.json"];
|
|
13
|
+
if ((await Promise.all(materializedFiles.map((name) => exists(path.join(options.evalDirectory, name))))).every(Boolean))
|
|
14
|
+
return null;
|
|
15
|
+
const failedResult = await readJSON(path.join(options.evalDirectory, "result.json"), null);
|
|
16
|
+
if (failedResult === null)
|
|
17
|
+
return null;
|
|
18
|
+
if (options.rerunType !== "candidate-restart" || options.selector.mode !== "invalid") {
|
|
19
|
+
throw unavailable("an eval with incomplete preparation can only be restarted with candidate-restart --invalid");
|
|
20
|
+
}
|
|
21
|
+
if (failedResult?.status === "cancelled") {
|
|
22
|
+
throw new HitchError("cancelled eval cannot be rerun", { code: "eval_rerun_cancelled", exitCode: 2 });
|
|
23
|
+
}
|
|
24
|
+
if (failedResult?.status !== "failed" || !Array.isArray(failedResult.trials) || failedResult.trials.length !== 0) {
|
|
25
|
+
throw unavailable("eval preparation is incomplete but the source is not a failed pre-execution attempt");
|
|
26
|
+
}
|
|
27
|
+
const taskIds = await resolveLocalDatasetTaskIds(request.dataset);
|
|
28
|
+
if (taskIds === null || taskIds.length === 0) {
|
|
29
|
+
throw unavailable("incomplete preparation recovery requires an enumerable local task plan");
|
|
30
|
+
}
|
|
31
|
+
const logicalPlan = await readEvalLogicalPlan(options.evalDirectory, options.evalId, request).catch((error) => {
|
|
32
|
+
throw unavailable(error.message);
|
|
33
|
+
});
|
|
34
|
+
let modelCapturePlan = logicalPlanModelCapture(logicalPlan) ?? options.modelCapturePlan;
|
|
35
|
+
if (!modelCapturePlan && !await exists(path.join(options.evalDirectory, "submission.json"))) {
|
|
36
|
+
modelCapturePlan = defaultModelCapturePlan();
|
|
37
|
+
}
|
|
38
|
+
if (!modelCapturePlan) {
|
|
39
|
+
throw unavailable("incomplete preparation recovery requires the original model capture plan");
|
|
40
|
+
}
|
|
41
|
+
await writeRequest(rerunDirectory, options, startedAt, [], []);
|
|
42
|
+
await writeState(statePath, options, startedAt, "running", [], [], [], []);
|
|
43
|
+
await archiveIncompleteAttempt(options, rerunDirectory, failedResult);
|
|
44
|
+
try {
|
|
45
|
+
const restarted = await runEval({
|
|
46
|
+
evalId: validateEvalId(options.evalId),
|
|
47
|
+
root: options.root,
|
|
48
|
+
request,
|
|
49
|
+
normalizedRequest: request,
|
|
50
|
+
precreated: true,
|
|
51
|
+
replaceTerminal: true,
|
|
52
|
+
env: options.env ?? process.env,
|
|
53
|
+
...(options.maxConcurrentOverride === undefined ? {} : { maxConcurrentOverride: options.maxConcurrentOverride }),
|
|
54
|
+
...(options.harborExecutable === undefined ? {} : { harborExecutable: options.harborExecutable }),
|
|
55
|
+
...(options.signal ? { signal: options.signal } : {}),
|
|
56
|
+
...(options.trialBundleGraceMs === undefined ? {} : { trialBundleGraceMs: options.trialBundleGraceMs }),
|
|
57
|
+
...(options.executionResources ? { executionResources: options.executionResources } : {}),
|
|
58
|
+
...(options.executionResourceSource ? { executionResourceSource: options.executionResourceSource } : {}),
|
|
59
|
+
executionStrategy: options.executionStrategy
|
|
60
|
+
?? (logicalPlan?.attempt_execution === "harbor-task-slots-v1" ? "local-task-slots-v1" : "legacy-attempt-shards"),
|
|
61
|
+
...(options.environmentBuildMode ? { environmentBuildMode: options.environmentBuildMode } : {}),
|
|
62
|
+
modelCapturePlan,
|
|
63
|
+
...(options.harborArtifactBuilder ? { harborArtifactBuilder: options.harborArtifactBuilder } : {}),
|
|
64
|
+
});
|
|
65
|
+
if (restarted.status === "cancelled") {
|
|
66
|
+
throw new HitchError("eval rerun was aborted", { code: "eval_rerun_aborted", exitCode: 9 });
|
|
67
|
+
}
|
|
68
|
+
const [planValue, executionPlanValue, progress] = await Promise.all([
|
|
69
|
+
readJSON(path.join(options.evalDirectory, "plan.json"), null),
|
|
70
|
+
readJSON(path.join(options.evalDirectory, "execution-plan.json"), null),
|
|
71
|
+
readEvalProgress(options.evalDirectory),
|
|
72
|
+
]);
|
|
73
|
+
if (planValue === null || executionPlanValue === null || progress === null)
|
|
74
|
+
throw restartFailure(restarted);
|
|
75
|
+
const plan = parseRecoveredPlan(planValue, options.evalId, request);
|
|
76
|
+
validateProgressPlan(progress, plan, request, options.evalId);
|
|
77
|
+
const selectedTrials = sortSlots(plan.tasks.flatMap((taskId) => Array.from({ length: plan.attempts }, (_, index) => ({ task_id: taskId, attempt: index + 1 }))));
|
|
78
|
+
const valid = new Set(progress.trials.filter((trial) => trial.observation_status === "valid").map(slotKey));
|
|
79
|
+
const repairedTrials = selectedTrials.filter((trial) => valid.has(slotKey(trial)));
|
|
80
|
+
const remainingTrials = invalidTrialSlots(plan.tasks, plan.attempts, progress);
|
|
81
|
+
const selectedTasks = uniqueTasks(selectedTrials);
|
|
82
|
+
const repairedTasks = uniqueTasks(repairedTrials);
|
|
83
|
+
const remainingTasks = uniqueTasks(remainingTrials);
|
|
84
|
+
const completedAt = new Date().toISOString();
|
|
85
|
+
await writeRequest(rerunDirectory, options, startedAt, selectedTasks, selectedTrials);
|
|
86
|
+
const output = {
|
|
87
|
+
schema_version: "1",
|
|
88
|
+
kind: "eval-rerun",
|
|
89
|
+
rerun_id: options.rerunId,
|
|
90
|
+
rerun_type: options.rerunType,
|
|
91
|
+
semantics: evalRerunSemantics(options.rerunType),
|
|
92
|
+
eval_id: options.evalId,
|
|
93
|
+
status: "completed",
|
|
94
|
+
selected_tasks: selectedTasks,
|
|
95
|
+
repaired_tasks: repairedTasks,
|
|
96
|
+
remaining_invalid_tasks: remainingTasks,
|
|
97
|
+
selected_trials: selectedTrials,
|
|
98
|
+
repaired_trials: repairedTrials,
|
|
99
|
+
remaining_invalid_trials: remainingTrials,
|
|
100
|
+
eval_status: restarted.status === "succeeded" ? "succeeded" : "failed",
|
|
101
|
+
started_at: startedAt,
|
|
102
|
+
completed_at: completedAt,
|
|
103
|
+
};
|
|
104
|
+
await writeState(statePath, options, startedAt, "completed", selectedTasks, selectedTrials, repairedTasks, repairedTrials, {
|
|
105
|
+
completedAt,
|
|
106
|
+
evalStatus: output.eval_status,
|
|
107
|
+
remainingTasks,
|
|
108
|
+
remainingTrials,
|
|
109
|
+
});
|
|
110
|
+
return output;
|
|
111
|
+
}
|
|
112
|
+
catch (error) {
|
|
113
|
+
await writeState(statePath, options, startedAt, "failed", [], [], [], [], {
|
|
114
|
+
completedAt: new Date().toISOString(),
|
|
115
|
+
errorCode: error instanceof HitchError ? error.code : "eval_rerun_failed",
|
|
116
|
+
}).catch(() => { });
|
|
117
|
+
throw error;
|
|
118
|
+
}
|
|
119
|
+
}
|
|
120
|
+
async function archiveIncompleteAttempt(options, rerunDirectory, result) {
|
|
121
|
+
const archive = path.join(rerunDirectory, "previous-attempt");
|
|
122
|
+
await ensureDir(archive);
|
|
123
|
+
for (const name of ["request.json", "submission.json", "control.json", "result.json"]) {
|
|
124
|
+
const source = path.join(options.evalDirectory, name);
|
|
125
|
+
if (await exists(source))
|
|
126
|
+
await cp(source, path.join(archive, name), { recursive: true, errorOnExist: true });
|
|
127
|
+
}
|
|
128
|
+
const stable = new Set(["request.json", "submission.json", "control.json", "result.json", "reruns"]);
|
|
129
|
+
for (const entry of await readdir(options.evalDirectory, { withFileTypes: true })) {
|
|
130
|
+
if (stable.has(entry.name))
|
|
131
|
+
continue;
|
|
132
|
+
await rename(path.join(options.evalDirectory, entry.name), path.join(archive, entry.name));
|
|
133
|
+
}
|
|
134
|
+
const failure = result.error && typeof result.error === "object" && !Array.isArray(result.error)
|
|
135
|
+
? result.error
|
|
136
|
+
: null;
|
|
137
|
+
await atomicWriteJSON(path.join(archive, "recovery.json"), {
|
|
138
|
+
schema_version: SCHEMA_VERSION,
|
|
139
|
+
kind: "eval-preparation-restart",
|
|
140
|
+
eval_id: options.evalId,
|
|
141
|
+
rerun_id: options.rerunId,
|
|
142
|
+
...(typeof result.failure_stage === "string" ? { failure_stage: result.failure_stage } : {}),
|
|
143
|
+
...(typeof failure?.code === "string" ? { failure_code: failure.code } : {}),
|
|
144
|
+
archived_at: new Date().toISOString(),
|
|
145
|
+
});
|
|
146
|
+
}
|
|
147
|
+
function restartFailure(result) {
|
|
148
|
+
const error = result.error && typeof result.error === "object" && !Array.isArray(result.error)
|
|
149
|
+
? result.error
|
|
150
|
+
: null;
|
|
151
|
+
const code = typeof error?.code === "string" && error.code ? error.code : "eval_rerun_preparation_failed";
|
|
152
|
+
const stage = typeof result.failure_stage === "string" ? result.failure_stage : "preparing";
|
|
153
|
+
const message = typeof error?.message === "string" && error.message ? error.message : "eval restart did not produce an executable plan";
|
|
154
|
+
return new HitchError(`eval restart failed during ${stage}: ${message}`, {
|
|
155
|
+
code,
|
|
156
|
+
exitCode: Number.isSafeInteger(result.exit_code) ? result.exit_code : 12,
|
|
157
|
+
});
|
|
158
|
+
}
|
|
159
|
+
function parseRecoveredPlan(value, evalId, request) {
|
|
160
|
+
if (!value || typeof value !== "object" || Array.isArray(value))
|
|
161
|
+
throw unavailable("recovered eval plan is missing");
|
|
162
|
+
const plan = value;
|
|
163
|
+
if (plan.schema_version !== SCHEMA_VERSION || plan.eval_id !== evalId || plan.dataset !== request.dataset
|
|
164
|
+
|| plan.benchmark_id !== request.benchmark_id || plan.benchmark_revision !== request.benchmark_revision
|
|
165
|
+
|| !Number.isSafeInteger(plan.attempts) || plan.attempts !== request.attempts) {
|
|
166
|
+
throw unavailable("recovered eval plan identity changed");
|
|
167
|
+
}
|
|
168
|
+
if (!Array.isArray(plan.tasks) || plan.tasks.length === 0
|
|
169
|
+
|| plan.tasks.some((task) => typeof task !== "string" || task.length === 0)
|
|
170
|
+
|| new Set(plan.tasks).size !== plan.tasks.length)
|
|
171
|
+
throw unavailable("recovered eval task plan is invalid");
|
|
172
|
+
return { tasks: [...plan.tasks], attempts: plan.attempts };
|
|
173
|
+
}
|
|
174
|
+
async function writeRequest(directory, options, startedAt, tasks, trials) {
|
|
175
|
+
await atomicWriteJSON(path.join(directory, "request.json"), {
|
|
176
|
+
schema_version: SCHEMA_VERSION,
|
|
177
|
+
rerun_id: options.rerunId,
|
|
178
|
+
eval_id: options.evalId,
|
|
179
|
+
rerun_type: options.rerunType,
|
|
180
|
+
semantics: evalRerunSemantics(options.rerunType),
|
|
181
|
+
mode: options.selector.mode,
|
|
182
|
+
tasks: [...tasks],
|
|
183
|
+
trials: sortSlots(trials),
|
|
184
|
+
base_generation: 0,
|
|
185
|
+
created_at: startedAt,
|
|
186
|
+
});
|
|
187
|
+
}
|
|
188
|
+
async function writeState(file, options, startedAt, status, tasks, trials, repairedTasks, repairedTrials, terminal = {}) {
|
|
189
|
+
await atomicWriteJSON(file, {
|
|
190
|
+
schema_version: SCHEMA_VERSION,
|
|
191
|
+
rerun_id: options.rerunId,
|
|
192
|
+
eval_id: options.evalId,
|
|
193
|
+
rerun_type: options.rerunType,
|
|
194
|
+
semantics: evalRerunSemantics(options.rerunType),
|
|
195
|
+
status,
|
|
196
|
+
tasks: [...tasks],
|
|
197
|
+
trials: sortSlots(trials),
|
|
198
|
+
repaired_tasks: [...repairedTasks],
|
|
199
|
+
repaired_trials: sortSlots(repairedTrials),
|
|
200
|
+
started_at: startedAt,
|
|
201
|
+
...(terminal.completedAt ? { completed_at: terminal.completedAt } : {}),
|
|
202
|
+
...(terminal.evalStatus ? { eval_status: terminal.evalStatus } : {}),
|
|
203
|
+
...(terminal.remainingTasks ? { remaining_invalid_tasks: [...terminal.remainingTasks] } : {}),
|
|
204
|
+
...(terminal.remainingTrials ? { remaining_invalid_trials: sortSlots(terminal.remainingTrials) } : {}),
|
|
205
|
+
...(terminal.errorCode ? { error: { code: terminal.errorCode } } : {}),
|
|
206
|
+
});
|
|
207
|
+
}
|
|
208
|
+
async function exists(file) {
|
|
209
|
+
try {
|
|
210
|
+
await lstat(file);
|
|
211
|
+
return true;
|
|
212
|
+
}
|
|
213
|
+
catch (error) {
|
|
214
|
+
if (error.code === "ENOENT")
|
|
215
|
+
return false;
|
|
216
|
+
throw error;
|
|
217
|
+
}
|
|
218
|
+
}
|
|
219
|
+
function unavailable(message) {
|
|
220
|
+
return new HitchError(message, { code: "eval_rerun_unavailable", exitCode: 2 });
|
|
221
|
+
}
|
|
222
|
+
function defaultModelCapturePlan() {
|
|
223
|
+
return { requested_mode: "native", effective_mode: "native", required: false };
|
|
224
|
+
}
|
|
225
|
+
//# sourceMappingURL=preparation-rerun.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"preparation-rerun.js","sourceRoot":"","sources":["../../../src/evals/preparation-rerun.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,kBAAkB,CAAC;AAC9D,OAAO,IAAI,MAAM,WAAW,CAAC;AAE7B,OAAO,EAAE,UAAU,EAAE,cAAc,EAAE,eAAe,EAAE,SAAS,EAAE,QAAQ,EAAE,MAAM,wBAAwB,CAAC;AAC1G,OAAO,EAAE,uBAAuB,EAAE,mBAAmB,EAAE,MAAM,wBAAwB,CAAC;AACtF,OAAO,EAAE,gBAAgB,EAAE,MAAM,eAAe,CAAC;AACjD,OAAO,EAAE,0BAA0B,EAAE,cAAc,EAAE,MAAM,cAAc,CAAC;AAC1E,OAAO,EAAE,kBAAkB,EAAE,MAAM,kBAAkB,CAAC;AAEtD,OAAO,EAAE,iBAAiB,EAAE,OAAO,EAAE,SAAS,EAAE,WAAW,EAAE,oBAAoB,EAAE,MAAM,kBAAkB,CAAC;AAE5G,OAAO,EAAE,OAAO,EAAE,MAAM,cAAc,CAAC;AASvC,qFAAqF;AACrF,MAAM,CAAC,KAAK,UAAU,qBAAqB,CACzC,OAAwB,EACxB,OAAoB,EACpB,SAAiB,EACjB,cAAsB,EACtB,SAAiB;IAEjB,MAAM,iBAAiB,GAAG,CAAC,WAAW,EAAE,qBAAqB,EAAE,eAAe,CAAC,CAAC;IAChF,IAAI,CAAC,MAAM,OAAO,CAAC,GAAG,CAAC,iBAAiB,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,aAAa,EAAE,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC;QAAE,OAAO,IAAI,CAAC;IACrI,MAAM,YAAY,GAAG,MAAM,QAAQ,CAAiC,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,aAAa,EAAE,aAAa,CAAC,EAAE,IAAI,CAAC,CAAC;IAC3H,IAAI,YAAY,KAAK,IAAI;QAAE,OAAO,IAAI,CAAC;IACvC,IAAI,OAAO,CAAC,SAAS,KAAK,mBAAmB,IAAI,OAAO,CAAC,QAAQ,CAAC,IAAI,KAAK,SAAS,EAAE,CAAC;QACrF,MAAM,WAAW,CAAC,4FAA4F,CAAC,CAAC;IAClH,CAAC;IACD,IAAI,YAAY,EAAE,MAAM,KAAK,WAAW,EAAE,CAAC;QACzC,MAAM,IAAI,UAAU,CAAC,gCAAgC,EAAE,EAAE,IAAI,EAAE,sBAAsB,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC,CAAC;IACxG,CAAC;IACD,IAAI,YAAY,EAAE,MAAM,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,YAAY,CAAC,MAAM,CAAC,IAAI,YAAY,CAAC,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACjH,MAAM,WAAW,CAAC,qFAAqF,CAAC,CAAC;IAC3G,CAAC;IACD,MAAM,OAAO,GAAG,MAAM,0BAA0B,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;IAClE,IAAI,OAAO,KAAK,IAAI,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC7C,MAAM,WAAW,CAAC,wEAAwE,CAAC,CAAC;IAC9F,CAAC;IAED,MAAM,WAAW,GAAG,MAAM,mBAAmB,CAAC,OAAO,CAAC,aAAa,EAAE,OAAO,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,KAAK,CAAC,CAAC,KAAK,EAAE,EAAE;QAC5G,MAAM,WAAW,CAAE,KAAe,CAAC,OAAO,CAAC,CAAC;IAC9C,CAAC,CAAC,CAAC;IACH,IAAI,gBAAgB,GAAG,uBAAuB,CAAC,WAAW,CAAC,IAAI,OAAO,CAAC,gBAAgB,CAAC;IACxF,IAAI,CAAC,gBAAgB,IAAI,CAAC,MAAM,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,aAAa,EAAE,iBAAiB,CAAC,CAAC,EAAE,CAAC;QAC5F,gBAAgB,GAAG,uBAAuB,EAAE,CAAC;IAC/C,CAAC;IACD,IAAI,CAAC,gBAAgB,EAAE,CAAC;QACtB,MAAM,WAAW,CAAC,0EAA0E,CAAC,CAAC;IAChG,CAAC;IACD,MAAM,YAAY,CAAC,cAAc,EAAE,OAAO,EAAE,SAAS,EAAE,EAAE,EAAE,EAAE,CAAC,CAAC;IAC/D,MAAM,UAAU,CAAC,SAAS,EAAE,OAAO,EAAE,SAAS,EAAE,SAAS,EAAE,EAAE,EAAE,EAAE,EAAE,EAAE,EAAE,EAAE,CAAC,CAAC;IAC3E,MAAM,wBAAwB,CAAC,OAAO,EAAE,cAAc,EAAE,YAAY,CAAC,CAAC;IAEtE,IAAI,CAAC;QACH,MAAM,SAAS,GAAG,MAAM,OAAO,CAAC;YAC9B,MAAM,EAAE,cAAc,CAAC,OAAO,CAAC,MAAM,CAAC;YACtC,IAAI,EAAE,OAAO,CAAC,IAAI;YAClB,OAAO;YACP,iBAAiB,EAAE,OAAO;YAC1B,UAAU,EAAE,IAAI;YAChB,eAAe,EAAE,IAAI;YACrB,GAAG,EAAE,OAAO,CAAC,GAAG,IAAI,OAAO,CAAC,GAAG;YAC/B,GAAG,CAAC,OAAO,CAAC,qBAAqB,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,qBAAqB,EAAE,OAAO,CAAC,qBAAqB,EAAE,CAAC;YAChH,GAAG,CAAC,OAAO,CAAC,gBAAgB,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,gBAAgB,EAAE,OAAO,CAAC,gBAAgB,EAAE,CAAC;YACjG,GAAG,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACrD,GAAG,CAAC,OAAO,CAAC,kBAAkB,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,kBAAkB,EAAE,OAAO,CAAC,kBAAkB,EAAE,CAAC;YACvG,GAAG,CAAC,OAAO,CAAC,kBAAkB,CAAC,CAAC,CAAC,EAAE,kBAAkB,EAAE,OAAO,CAAC,kBAAkB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACzF,GAAG,CAAC,OAAO,CAAC,uBAAuB,CAAC,CAAC,CAAC,EAAE,uBAAuB,EAAE,OAAO,CAAC,uBAAuB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACxG,iBAAiB,EAAE,OAAO,CAAC,iBAAiB;mBACvC,CAAC,WAAW,EAAE,iBAAiB,KAAK,sBAAsB,CAAC,CAAC,CAAC,qBAAqB,CAAC,CAAC,CAAC,uBAAuB,CAAC;YAClH,GAAG,CAAC,OAAO,CAAC,oBAAoB,CAAC,CAAC,CAAC,EAAE,oBAAoB,EAAE,OAAO,CAAC,oBAAoB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YAC/F,gBAAgB;YAChB,GAAG,CAAC,OAAO,CAAC,qBAAqB,CAAC,CAAC,CAAC,EAAE,qBAAqB,EAAE,OAAO,CAAC,qBAAqB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SACnG,CAAC,CAAC;QACH,IAAI,SAAS,CAAC,MAAM,KAAK,WAAW,EAAE,CAAC;YACrC,MAAM,IAAI,UAAU,CAAC,wBAAwB,EAAE,EAAE,IAAI,EAAE,oBAAoB,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC,CAAC;QAC9F,CAAC;QACD,MAAM,CAAC,SAAS,EAAE,kBAAkB,EAAE,QAAQ,CAAC,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;YAClE,QAAQ,CAAiB,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,aAAa,EAAE,WAAW,CAAC,EAAE,IAAI,CAAC;YAC7E,QAAQ,CAAiB,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,aAAa,EAAE,qBAAqB,CAAC,EAAE,IAAI,CAAC;YACvF,gBAAgB,CAAC,OAAO,CAAC,aAAa,CAAC;SACxC,CAAC,CAAC;QACH,IAAI,SAAS,KAAK,IAAI,IAAI,kBAAkB,KAAK,IAAI,IAAI,QAAQ,KAAK,IAAI;YAAE,MAAM,cAAc,CAAC,SAAS,CAAC,CAAC;QAE5G,MAAM,IAAI,GAAG,kBAAkB,CAAC,SAAS,EAAE,OAAO,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;QACpE,oBAAoB,CAAC,QAAQ,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC;QAC9D,MAAM,cAAc,GAAG,SAAS,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,CACxE,EAAE,MAAM,EAAE,IAAI,CAAC,QAAQ,EAAE,EACzB,CAAC,CAAC,EAAE,KAAK,EAAE,EAAE,CAAC,CAAC,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,KAAK,GAAG,CAAC,EAAE,CAAC,CACxD,CAAC,CAAC,CAAC;QACJ,MAAM,KAAK,GAAG,IAAI,GAAG,CAAC,QAAQ,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,kBAAkB,KAAK,OAAO,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC;QAC5G,MAAM,cAAc,GAAG,cAAc,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,GAAG,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;QACnF,MAAM,eAAe,GAAG,iBAAiB,CAAC,IAAI,CAAC,KAAK,EAAE,IAAI,CAAC,QAAQ,EAAE,QAAQ,CAAC,CAAC;QAC/E,MAAM,aAAa,GAAG,WAAW,CAAC,cAAc,CAAC,CAAC;QAClD,MAAM,aAAa,GAAG,WAAW,CAAC,cAAc,CAAC,CAAC;QAClD,MAAM,cAAc,GAAG,WAAW,CAAC,eAAe,CAAC,CAAC;QACpD,MAAM,WAAW,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;QAC7C,MAAM,YAAY,CAAC,cAAc,EAAE,OAAO,EAAE,SAAS,EAAE,aAAa,EAAE,cAAc,CAAC,CAAC;QACtF,MAAM,MAAM,GAAoB;YAC9B,cAAc,EAAE,GAAG;YACnB,IAAI,EAAE,YAAY;YAClB,QAAQ,EAAE,OAAO,CAAC,OAAO;YACzB,UAAU,EAAE,OAAO,CAAC,SAAS;YAC7B,SAAS,EAAE,kBAAkB,CAAC,OAAO,CAAC,SAAS,CAAC;YAChD,OAAO,EAAE,OAAO,CAAC,MAAM;YACvB,MAAM,EAAE,WAAW;YACnB,cAAc,EAAE,aAAa;YAC7B,cAAc,EAAE,aAAa;YAC7B,uBAAuB,EAAE,cAAc;YACvC,eAAe,EAAE,cAAc;YAC/B,eAAe,EAAE,cAAc;YAC/B,wBAAwB,EAAE,eAAe;YACzC,WAAW,EAAE,SAAS,CAAC,MAAM,KAAK,WAAW,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,QAAQ;YACtE,UAAU,EAAE,SAAS;YACrB,YAAY,EAAE,WAAW;SAC1B,CAAC;QACF,MAAM,UAAU,CAAC,SAAS,EAAE,OAAO,EAAE,SAAS,EAAE,WAAW,EAAE,aAAa,EAAE,cAAc,EAAE,aAAa,EAAE,cAAc,EAAE;YACzH,WAAW;YACX,UAAU,EAAE,MAAM,CAAC,WAAW;YAC9B,cAAc;YACd,eAAe;SAChB,CAAC,CAAC;QACH,OAAO,MAAM,CAAC;IAChB,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,MAAM,UAAU,CAAC,SAAS,EAAE,OAAO,EAAE,SAAS,EAAE,QAAQ,EAAE,EAAE,EAAE,EAAE,EAAE,EAAE,EAAE,EAAE,EAAE;YACxE,WAAW,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;YACrC,SAAS,EAAE,KAAK,YAAY,UAAU,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,mBAAmB;SAC1E,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,CAAC;QACnB,MAAM,KAAK,CAAC;IACd,CAAC;AACH,CAAC;AAED,KAAK,UAAU,wBAAwB,CACrC,OAAwB,EACxB,cAAsB,EACtB,MAA+B;IAE/B,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,kBAAkB,CAAC,CAAC;IAC9D,MAAM,SAAS,CAAC,OAAO,CAAC,CAAC;IACzB,KAAK,MAAM,IAAI,IAAI,CAAC,cAAc,EAAE,iBAAiB,EAAE,cAAc,EAAE,aAAa,CAAC,EAAE,CAAC;QACtF,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,aAAa,EAAE,IAAI,CAAC,CAAC;QACtD,IAAI,MAAM,MAAM,CAAC,MAAM,CAAC;YAAE,MAAM,EAAE,CAAC,MAAM,EAAE,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,IAAI,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,CAAC,CAAC;IAChH,CAAC;IACD,MAAM,MAAM,GAAG,IAAI,GAAG,CAAC,CAAC,cAAc,EAAE,iBAAiB,EAAE,cAAc,EAAE,aAAa,EAAE,QAAQ,CAAC,CAAC,CAAC;IACrG,KAAK,MAAM,KAAK,IAAI,MAAM,OAAO,CAAC,OAAO,CAAC,aAAa,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,EAAE,CAAC;QAClF,IAAI,MAAM,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC;YAAE,SAAS;QACrC,MAAM,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,aAAa,EAAE,KAAK,CAAC,IAAI,CAAC,EAAE,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC;IAC7F,CAAC;IACD,MAAM,OAAO,GAAG,MAAM,CAAC,KAAK,IAAI,OAAO,MAAM,CAAC,KAAK,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC;QAC9F,CAAC,CAAC,MAAM,CAAC,KAAgC;QACzC,CAAC,CAAC,IAAI,CAAC;IACT,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,eAAe,CAAC,EAAE;QACzD,cAAc,EAAE,cAAc;QAC9B,IAAI,EAAE,0BAA0B;QAChC,OAAO,EAAE,OAAO,CAAC,MAAM;QACvB,QAAQ,EAAE,OAAO,CAAC,OAAO;QACzB,GAAG,CAAC,OAAO,MAAM,CAAC,aAAa,KAAK,QAAQ,CAAC,CAAC,CAAC,EAAE,aAAa,EAAE,MAAM,CAAC,aAAa,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC5F,GAAG,CAAC,OAAO,OAAO,EAAE,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,EAAE,YAAY,EAAE,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC5E,WAAW,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;KACtC,CAAC,CAAC;AACL,CAAC;AAED,SAAS,cAAc,CAAC,MAA+B;IACrD,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,IAAI,OAAO,MAAM,CAAC,KAAK,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC;QAC5F,CAAC,CAAC,MAAM,CAAC,KAAgC;QACzC,CAAC,CAAC,IAAI,CAAC;IACT,MAAM,IAAI,GAAG,OAAO,KAAK,EAAE,IAAI,KAAK,QAAQ,IAAI,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,+BAA+B,CAAC;IAC1G,MAAM,KAAK,GAAG,OAAO,MAAM,CAAC,aAAa,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,CAAC,WAAW,CAAC;IAC5F,MAAM,OAAO,GAAG,OAAO,KAAK,EAAE,OAAO,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,iDAAiD,CAAC;IACxI,OAAO,IAAI,UAAU,CAAC,8BAA8B,KAAK,KAAK,OAAO,EAAE,EAAE;QACvE,IAAI;QACJ,QAAQ,EAAE,MAAM,CAAC,aAAa,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,SAAmB,CAAC,CAAC,CAAC,EAAE;KACnF,CAAC,CAAC;AACL,CAAC;AAED,SAAS,kBAAkB,CAAC,KAAc,EAAE,MAAc,EAAE,OAAoB;IAC9E,IAAI,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;QAAE,MAAM,WAAW,CAAC,gCAAgC,CAAC,CAAC;IACrH,MAAM,IAAI,GAAG,KAAgC,CAAC;IAC9C,IAAI,IAAI,CAAC,cAAc,KAAK,cAAc,IAAI,IAAI,CAAC,OAAO,KAAK,MAAM,IAAI,IAAI,CAAC,OAAO,KAAK,OAAO,CAAC,OAAO;WACpG,IAAI,CAAC,YAAY,KAAK,OAAO,CAAC,YAAY,IAAI,IAAI,CAAC,kBAAkB,KAAK,OAAO,CAAC,kBAAkB;WACpG,CAAC,MAAM,CAAC,aAAa,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,IAAI,CAAC,QAAQ,KAAK,OAAO,CAAC,QAAQ,EAAE,CAAC;QAChF,MAAM,WAAW,CAAC,sCAAsC,CAAC,CAAC;IAC5D,CAAC;IACD,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,IAAI,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC;WACpD,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,OAAO,IAAI,KAAK,QAAQ,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC,CAAC;WACxE,IAAI,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,IAAI,KAAK,IAAI,CAAC,KAAK,CAAC,MAAM;QAAE,MAAM,WAAW,CAAC,qCAAqC,CAAC,CAAC;IAC9G,OAAO,EAAE,KAAK,EAAE,CAAC,GAAG,IAAI,CAAC,KAAiB,CAAC,EAAE,QAAQ,EAAE,IAAI,CAAC,QAAkB,EAAE,CAAC;AACnF,CAAC;AAED,KAAK,UAAU,YAAY,CACzB,SAAiB,EACjB,OAAwB,EACxB,SAAiB,EACjB,KAAwB,EACxB,MAAgC;IAEhC,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,cAAc,CAAC,EAAE;QAC1D,cAAc,EAAE,cAAc;QAC9B,QAAQ,EAAE,OAAO,CAAC,OAAO;QACzB,OAAO,EAAE,OAAO,CAAC,MAAM;QACvB,UAAU,EAAE,OAAO,CAAC,SAAS;QAC7B,SAAS,EAAE,kBAAkB,CAAC,OAAO,CAAC,SAAS,CAAC;QAChD,IAAI,EAAE,OAAO,CAAC,QAAQ,CAAC,IAAI;QAC3B,KAAK,EAAE,CAAC,GAAG,KAAK,CAAC;QACjB,MAAM,EAAE,SAAS,CAAC,MAAM,CAAC;QACzB,eAAe,EAAE,CAAC;QAClB,UAAU,EAAE,SAAS;KACtB,CAAC,CAAC;AACL,CAAC;AAED,KAAK,UAAU,UAAU,CACvB,IAAY,EACZ,OAAwB,EACxB,SAAiB,EACjB,MAA0C,EAC1C,KAAwB,EACxB,MAAgC,EAChC,aAAgC,EAChC,cAAwC,EACxC,WAMI,EAAE;IAEN,MAAM,eAAe,CAAC,IAAI,EAAE;QAC1B,cAAc,EAAE,cAAc;QAC9B,QAAQ,EAAE,OAAO,CAAC,OAAO;QACzB,OAAO,EAAE,OAAO,CAAC,MAAM;QACvB,UAAU,EAAE,OAAO,CAAC,SAAS;QAC7B,SAAS,EAAE,kBAAkB,CAAC,OAAO,CAAC,SAAS,CAAC;QAChD,MAAM;QACN,KAAK,EAAE,CAAC,GAAG,KAAK,CAAC;QACjB,MAAM,EAAE,SAAS,CAAC,MAAM,CAAC;QACzB,cAAc,EAAE,CAAC,GAAG,aAAa,CAAC;QAClC,eAAe,EAAE,SAAS,CAAC,cAAc,CAAC;QAC1C,UAAU,EAAE,SAAS;QACrB,GAAG,CAAC,QAAQ,CAAC,WAAW,CAAC,CAAC,CAAC,EAAE,YAAY,EAAE,QAAQ,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACvE,GAAG,CAAC,QAAQ,CAAC,UAAU,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,QAAQ,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACpE,GAAG,CAAC,QAAQ,CAAC,cAAc,CAAC,CAAC,CAAC,EAAE,uBAAuB,EAAE,CAAC,GAAG,QAAQ,CAAC,cAAc,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC7F,GAAG,CAAC,QAAQ,CAAC,eAAe,CAAC,CAAC,CAAC,EAAE,wBAAwB,EAAE,SAAS,CAAC,QAAQ,CAAC,eAAe,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACtG,GAAG,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,QAAQ,CAAC,SAAS,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KACvE,CAAC,CAAC;AACL,CAAC;AAED,KAAK,UAAU,MAAM,CAAC,IAAY;IAChC,IAAI,CAAC;QACH,MAAM,KAAK,CAAC,IAAI,CAAC,CAAC;QAClB,OAAO,IAAI,CAAC;IACd,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAK,KAA+B,CAAC,IAAI,KAAK,QAAQ;YAAE,OAAO,KAAK,CAAC;QACrE,MAAM,KAAK,CAAC;IACd,CAAC;AACH,CAAC;AAED,SAAS,WAAW,CAAC,OAAe;IAClC,OAAO,IAAI,UAAU,CAAC,OAAO,EAAE,EAAE,IAAI,EAAE,wBAAwB,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC,CAAC;AAClF,CAAC;AAED,SAAS,uBAAuB;IAC9B,OAAO,EAAE,cAAc,EAAE,QAAQ,EAAE,cAAc,EAAE,QAAQ,EAAE,QAAQ,EAAE,KAAK,EAAE,CAAC;AACjF,CAAC"}
|
|
@@ -39,8 +39,8 @@ export function mergeEvalProgressTrial(progress, trial, now = new Date().toISOSt
|
|
|
39
39
|
throw new TypeError(`eval progress trial identity conflict: ${parsed.trial_id}`);
|
|
40
40
|
return progress;
|
|
41
41
|
}
|
|
42
|
-
if (progress.trials.some((item) => item
|
|
43
|
-
throw new TypeError(`eval progress run identity conflict: ${parsed
|
|
42
|
+
if (progress.trials.some((item) => evalTrialCandidateKey(item) === evalTrialCandidateKey(parsed))) {
|
|
43
|
+
throw new TypeError(`eval progress run identity conflict: ${evalTrialCandidateKey(parsed)}`);
|
|
44
44
|
}
|
|
45
45
|
if (progress.trials.some((item) => evalTrialKey(item) === evalTrialKey(parsed))) {
|
|
46
46
|
throw new TypeError(`eval progress logical trial conflict: ${parsed.task_id} attempt ${parsed.attempt}`);
|
|
@@ -64,6 +64,9 @@ export function mergeEvalProgressTrial(progress, trial, now = new Date().toISOSt
|
|
|
64
64
|
export function evalTrialKey(trial) {
|
|
65
65
|
return `${trial.task_id}\u0000${trial.attempt}`;
|
|
66
66
|
}
|
|
67
|
+
export function evalTrialCandidateKey(trial) {
|
|
68
|
+
return trial.run_group ? trial.run_group.run_group_id : trial.run_id;
|
|
69
|
+
}
|
|
67
70
|
/** Replace one invalid logical trial, or fill a missing slot, with a valid verifier result. */
|
|
68
71
|
export function replaceInvalidEvalProgressTrial(progress, trial, now = new Date().toISOString()) {
|
|
69
72
|
const parsed = parseEvalTrialRef(trial, "eval rerun trial");
|
|
@@ -76,8 +79,8 @@ export function replaceInvalidEvalProgressTrial(progress, trial, now = new Date(
|
|
|
76
79
|
return progress;
|
|
77
80
|
throw new TypeError(`eval rerun cannot replace valid task: ${parsed.task_id}`);
|
|
78
81
|
}
|
|
79
|
-
if (progress.trials.some((item) => item
|
|
80
|
-
throw new TypeError(`eval progress run identity conflict: ${parsed
|
|
82
|
+
if (progress.trials.some((item) => evalTrialCandidateKey(item) === evalTrialCandidateKey(parsed) && evalTrialKey(item) !== key)) {
|
|
83
|
+
throw new TypeError(`eval progress run identity conflict: ${evalTrialCandidateKey(parsed)}`);
|
|
81
84
|
}
|
|
82
85
|
if (progress.trials.some((item) => item.trial_id === parsed.trial_id && evalTrialKey(item) !== key)) {
|
|
83
86
|
throw new TypeError(`eval progress trial identity conflict: ${parsed.trial_id}`);
|
|
@@ -120,7 +123,7 @@ export function parseEvalProgress(value) {
|
|
|
120
123
|
throw new TypeError("eval progress trials are invalid");
|
|
121
124
|
const trials = record.trials.map((trial, index) => parseEvalTrialRef(trial, `eval progress trial ${index}`));
|
|
122
125
|
if (new Set(trials.map((trial) => trial.trial_id)).size !== trials.length
|
|
123
|
-
|| new Set(trials.map(
|
|
126
|
+
|| new Set(trials.map(evalTrialCandidateKey)).size !== trials.length)
|
|
124
127
|
throw new TypeError("eval progress trial identities are duplicated");
|
|
125
128
|
const sorted = [...trials].sort((left, right) => left.task_id.localeCompare(right.task_id)
|
|
126
129
|
|| left.attempt - right.attempt
|
|
@@ -169,7 +172,6 @@ export function parseEvalTrialRef(value, label = "eval trial") {
|
|
|
169
172
|
throw new TypeError(`${label} must be an object`);
|
|
170
173
|
const trial = value;
|
|
171
174
|
if (typeof trial.trial_id !== "string" || !trial.trial_id
|
|
172
|
-
|| typeof trial.run_id !== "string" || !RUN_ID.test(trial.run_id)
|
|
173
175
|
|| typeof trial.task_id !== "string" || !trial.task_id
|
|
174
176
|
|| !Number.isSafeInteger(trial.attempt) || trial.attempt < 1
|
|
175
177
|
|| (trial.observation_status !== "valid" && trial.observation_status !== "invalid")) {
|
|
@@ -178,21 +180,59 @@ export function parseEvalTrialRef(value, label = "eval trial") {
|
|
|
178
180
|
if (trial.observation_status === "valid" && (typeof trial.reward !== "number" || !Number.isFinite(trial.reward))) {
|
|
179
181
|
throw new TypeError(`${label} valid reward is invalid`);
|
|
180
182
|
}
|
|
183
|
+
const scores = trial.scores === undefined ? undefined : parseScores(trial.scores, label);
|
|
184
|
+
if (scores !== undefined && (trial.observation_status !== "valid" || scores.total_score !== trial.reward)) {
|
|
185
|
+
throw new TypeError(`${label} scores do not match the valid reward`);
|
|
186
|
+
}
|
|
181
187
|
if (trial.observation_status === "invalid" && (typeof trial.invalid_reason !== "string" || !trial.invalid_reason)) {
|
|
182
188
|
throw new TypeError(`${label} invalid reason is missing`);
|
|
183
189
|
}
|
|
184
190
|
if (trial.verifier_result_ref !== undefined && (typeof trial.verifier_result_ref !== "string" || !trial.verifier_result_ref)) {
|
|
185
191
|
throw new TypeError(`${label} verifier ref is invalid`);
|
|
186
192
|
}
|
|
187
|
-
|
|
193
|
+
const assessment = trial.assessment;
|
|
194
|
+
if (assessment !== undefined && (!assessment || typeof assessment.id !== "string" || !/^assessment_[a-f0-9]{32}$/.test(assessment.id)
|
|
195
|
+
|| typeof assessment.digest !== "string" || !/^sha256:[a-f0-9]{64}$/.test(assessment.digest)))
|
|
196
|
+
throw new TypeError(`${label} assessment is invalid`);
|
|
197
|
+
const group = trial.run_group;
|
|
198
|
+
if (group !== undefined) {
|
|
199
|
+
if (!group || typeof group !== "object" || Array.isArray(group) || Object.keys(group).some(key => !["run_group_id", "digest"].includes(key))
|
|
200
|
+
|| typeof group.run_group_id !== "string" || !/^run_group_[a-f0-9]{32}$/.test(group.run_group_id)
|
|
201
|
+
|| typeof group.digest !== "string" || !/^sha256:[a-f0-9]{64}$/.test(group.digest) || trial.run_id !== undefined || !assessment) {
|
|
202
|
+
throw new TypeError(`${label} phase group identity is invalid`);
|
|
203
|
+
}
|
|
204
|
+
}
|
|
205
|
+
else if (typeof trial.run_id !== "string" || !RUN_ID.test(trial.run_id))
|
|
206
|
+
throw new TypeError(`${label} run identity is invalid`);
|
|
207
|
+
const common = {
|
|
188
208
|
trial_id: trial.trial_id,
|
|
189
|
-
run_id: trial.run_id,
|
|
190
209
|
task_id: trial.task_id,
|
|
191
210
|
attempt: trial.attempt,
|
|
192
211
|
observation_status: trial.observation_status,
|
|
193
212
|
...(trial.reward === undefined ? {} : { reward: trial.reward }),
|
|
213
|
+
...(scores === undefined ? {} : { scores }),
|
|
194
214
|
...(trial.verifier_result_ref === undefined ? {} : { verifier_result_ref: trial.verifier_result_ref }),
|
|
195
215
|
...(trial.invalid_reason === undefined ? {} : { invalid_reason: trial.invalid_reason }),
|
|
196
216
|
};
|
|
217
|
+
const reference = assessment ? { id: assessment.id, digest: assessment.digest } : undefined;
|
|
218
|
+
return group ? { ...common, run_group: { run_group_id: group.run_group_id, digest: group.digest }, assessment: reference }
|
|
219
|
+
: { ...common, run_id: trial.run_id, ...(reference ? { assessment: reference } : {}) };
|
|
220
|
+
}
|
|
221
|
+
function parseScores(value, label) {
|
|
222
|
+
if (!value || typeof value !== "object" || Array.isArray(value))
|
|
223
|
+
throw new TypeError(`${label} scores are invalid`);
|
|
224
|
+
const record = value;
|
|
225
|
+
if (Object.keys(record).some((key) => !["total_score", "process_score", "normalization"].includes(key))
|
|
226
|
+
|| typeof record.total_score !== "number" || !Number.isFinite(record.total_score)
|
|
227
|
+
|| record.process_score !== undefined && (typeof record.process_score !== "number" || !Number.isFinite(record.process_score))
|
|
228
|
+
|| record.normalization !== "standard" && record.normalization !== "legacy-reward"
|
|
229
|
+
|| record.normalization === "legacy-reward" && record.process_score !== undefined) {
|
|
230
|
+
throw new TypeError(`${label} scores are invalid`);
|
|
231
|
+
}
|
|
232
|
+
return {
|
|
233
|
+
total_score: record.total_score,
|
|
234
|
+
...(record.process_score === undefined ? {} : { process_score: record.process_score }),
|
|
235
|
+
normalization: record.normalization,
|
|
236
|
+
};
|
|
197
237
|
}
|
|
198
238
|
//# sourceMappingURL=progress.js.map
|