agent-hitch 0.2.6 → 0.2.7
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 +14 -1
- package/dist/scripts/canary-buildkit-secrets.js +166 -0
- package/dist/scripts/canary-buildkit-secrets.js.map +1 -0
- package/dist/scripts/canary-collision-domains.js +77 -0
- package/dist/scripts/canary-collision-domains.js.map +1 -0
- package/dist/scripts/canary-docker-image-observation.js +181 -0
- package/dist/scripts/canary-docker-image-observation.js.map +1 -0
- package/dist/scripts/canary-gpu-hardware.js +95 -0
- package/dist/scripts/canary-gpu-hardware.js.map +1 -0
- package/dist/scripts/canary-harbor-load.js +365 -0
- package/dist/scripts/canary-harbor-load.js.map +1 -0
- package/dist/scripts/canary-packaged-harness.js +75 -0
- package/dist/scripts/canary-packaged-harness.js.map +1 -0
- package/dist/scripts/canary-resource-load.js +188 -0
- package/dist/scripts/canary-resource-load.js.map +1 -0
- package/dist/scripts/check-architecture.js +8 -4
- package/dist/scripts/check-architecture.js.map +1 -1
- package/dist/scripts/docker-reaper-crash-child.js +21 -0
- package/dist/scripts/docker-reaper-crash-child.js.map +1 -0
- package/dist/src/adapters/catalog.js +1 -0
- package/dist/src/adapters/catalog.js.map +1 -1
- package/dist/src/adapters/discovery.js +1 -1
- package/dist/src/adapters/discovery.js.map +1 -1
- package/dist/src/adapters/providers/claude.js +6 -0
- package/dist/src/adapters/providers/claude.js.map +1 -1
- package/dist/src/adapters/providers/codex.js +6 -0
- package/dist/src/adapters/providers/codex.js.map +1 -1
- package/dist/src/adapters/providers/deepseek.js +6 -0
- package/dist/src/adapters/providers/deepseek.js.map +1 -1
- package/dist/src/adapters/providers/opencode.js +6 -0
- package/dist/src/adapters/providers/opencode.js.map +1 -1
- package/dist/src/adapters/providers/pi.js +6 -0
- package/dist/src/adapters/providers/pi.js.map +1 -1
- package/dist/src/artifacts/index.js +1 -0
- package/dist/src/artifacts/index.js.map +1 -1
- package/dist/src/backends/harbor/backend.js +53 -103
- package/dist/src/backends/harbor/backend.js.map +1 -1
- package/dist/src/backends/harbor/environment-config.js +92 -0
- package/dist/src/backends/harbor/environment-config.js.map +1 -0
- package/dist/src/backends/harbor/index.js +5 -1
- package/dist/src/backends/harbor/index.js.map +1 -1
- package/dist/src/backends/harbor/model-proxy-config.js +23 -0
- package/dist/src/backends/harbor/model-proxy-config.js.map +1 -0
- package/dist/src/backends/harbor/process.js +177 -0
- package/dist/src/backends/harbor/process.js.map +1 -0
- package/dist/src/backends/harbor/runtime-toolchain.js +30 -0
- package/dist/src/backends/harbor/runtime-toolchain.js.map +1 -0
- package/dist/src/backends/harbor/supervisor.js +74 -0
- package/dist/src/backends/harbor/supervisor.js.map +1 -0
- package/dist/src/backends/harbor/task-resources.js +206 -0
- package/dist/src/backends/harbor/task-resources.js.map +1 -0
- package/dist/src/backends/harbor/tools.js +6 -29
- package/dist/src/backends/harbor/tools.js.map +1 -1
- package/dist/src/backends/index.js +4 -1
- package/dist/src/backends/index.js.map +1 -1
- package/dist/src/cli/commands/daemon.js +135 -8
- package/dist/src/cli/commands/daemon.js.map +1 -1
- package/dist/src/cli/commands/eval.js +214 -4
- package/dist/src/cli/commands/eval.js.map +1 -1
- package/dist/src/cli/commands/images.js +40 -0
- package/dist/src/cli/commands/images.js.map +1 -0
- package/dist/src/cli/commands/run.js +6 -3
- package/dist/src/cli/commands/run.js.map +1 -1
- package/dist/src/cli/commands/runs.js +25 -3
- package/dist/src/cli/commands/runs.js.map +1 -1
- package/dist/src/cli/commands/worker.js +79 -0
- package/dist/src/cli/commands/worker.js.map +1 -0
- package/dist/src/cli/main.js +4 -0
- package/dist/src/cli/main.js.map +1 -1
- package/dist/src/cli/output.js +71 -3
- package/dist/src/cli/output.js.map +1 -1
- package/dist/src/control-plane/build-admission.js +69 -0
- package/dist/src/control-plane/build-admission.js.map +1 -0
- package/dist/src/control-plane/build-records.js +6 -0
- package/dist/src/control-plane/build-records.js.map +1 -0
- package/dist/src/control-plane/collisions.js +40 -0
- package/dist/src/control-plane/collisions.js.map +1 -0
- package/dist/src/control-plane/eval-control-state.js +22 -0
- package/dist/src/control-plane/eval-control-state.js.map +1 -0
- package/dist/src/control-plane/eval-control-work.js +39 -0
- package/dist/src/control-plane/eval-control-work.js.map +1 -0
- package/dist/src/control-plane/eval-image-resolution.js +52 -0
- package/dist/src/control-plane/eval-image-resolution.js.map +1 -0
- package/dist/src/control-plane/eval-image-services.js +30 -0
- package/dist/src/control-plane/eval-image-services.js.map +1 -0
- package/dist/src/control-plane/eval-records.js +252 -0
- package/dist/src/control-plane/eval-records.js.map +1 -0
- package/dist/src/control-plane/eval-recovery.js +181 -0
- package/dist/src/control-plane/eval-recovery.js.map +1 -0
- package/dist/src/control-plane/eval-scheduler.js +447 -0
- package/dist/src/control-plane/eval-scheduler.js.map +1 -0
- package/dist/src/control-plane/index.js +20 -0
- package/dist/src/control-plane/index.js.map +1 -0
- package/dist/src/control-plane/local-worker.js +49 -0
- package/dist/src/control-plane/local-worker.js.map +1 -0
- package/dist/src/control-plane/model-capture-planning.js +12 -0
- package/dist/src/control-plane/model-capture-planning.js.map +1 -0
- package/dist/src/control-plane/remote-result-transport.js +172 -0
- package/dist/src/control-plane/remote-result-transport.js.map +1 -0
- package/dist/src/control-plane/remote-work-coordinator.js +332 -0
- package/dist/src/control-plane/remote-work-coordinator.js.map +1 -0
- package/dist/src/control-plane/remote-work-inputs.js +266 -0
- package/dist/src/control-plane/remote-work-inputs.js.map +1 -0
- package/dist/src/control-plane/remote-work-recovery.js +181 -0
- package/dist/src/control-plane/remote-work-recovery.js.map +1 -0
- package/dist/src/control-plane/remote-worker-artifacts.js +135 -0
- package/dist/src/control-plane/remote-worker-artifacts.js.map +1 -0
- package/dist/src/control-plane/remote-worker-client.js +202 -0
- package/dist/src/control-plane/remote-worker-client.js.map +1 -0
- package/dist/src/control-plane/remote-worker-configuration.js +34 -0
- package/dist/src/control-plane/remote-worker-configuration.js.map +1 -0
- package/dist/src/control-plane/remote-worker-credentials.js +58 -0
- package/dist/src/control-plane/remote-worker-credentials.js.map +1 -0
- package/dist/src/control-plane/remote-worker-protocol.js +510 -0
- package/dist/src/control-plane/remote-worker-protocol.js.map +1 -0
- package/dist/src/control-plane/remote-worker-runner.js +214 -0
- package/dist/src/control-plane/remote-worker-runner.js.map +1 -0
- package/dist/src/control-plane/remote-workers.js +377 -0
- package/dist/src/control-plane/remote-workers.js.map +1 -0
- package/dist/src/control-plane/rerun-scheduler.js +411 -0
- package/dist/src/control-plane/rerun-scheduler.js.map +1 -0
- package/dist/src/control-plane/rerun-submission.js +56 -0
- package/dist/src/control-plane/rerun-submission.js.map +1 -0
- package/dist/src/control-plane/resources.js +182 -0
- package/dist/src/control-plane/resources.js.map +1 -0
- package/dist/src/control-plane/scheduler-eval-entry.js +32 -0
- package/dist/src/control-plane/scheduler-eval-entry.js.map +1 -0
- package/dist/src/control-plane/synthetic-result.js +21 -0
- package/dist/src/control-plane/synthetic-result.js.map +1 -0
- package/dist/src/control-plane/work-admission.js +21 -0
- package/dist/src/control-plane/work-admission.js.map +1 -0
- package/dist/src/control-plane/work-dispatcher.js +229 -0
- package/dist/src/control-plane/work-dispatcher.js.map +1 -0
- package/dist/src/controller-runtime/hash.js +12 -1
- package/dist/src/controller-runtime/hash.js.map +1 -1
- package/dist/src/controller-runtime/index.js +1 -1
- package/dist/src/controller-runtime/index.js.map +1 -1
- package/dist/src/controller-runtime/store.js +15 -6
- package/dist/src/controller-runtime/store.js.map +1 -1
- package/dist/src/daemon/client.js.map +1 -1
- package/dist/src/daemon/event-stream.js +46 -0
- package/dist/src/daemon/event-stream.js.map +1 -0
- package/dist/src/daemon/health.js +66 -0
- package/dist/src/daemon/health.js.map +1 -0
- package/dist/src/daemon/launcher.js +28 -1
- package/dist/src/daemon/launcher.js.map +1 -1
- package/dist/src/daemon/logging.js +23 -0
- package/dist/src/daemon/logging.js.map +1 -0
- package/dist/src/daemon/scheduler.js +26 -5
- package/dist/src/daemon/scheduler.js.map +1 -1
- package/dist/src/daemon/server.js +255 -98
- package/dist/src/daemon/server.js.map +1 -1
- package/dist/src/daemon/telemetry.js +163 -0
- package/dist/src/daemon/telemetry.js.map +1 -0
- package/dist/src/daemon/worker-routes.js +219 -0
- package/dist/src/daemon/worker-routes.js.map +1 -0
- package/dist/src/domain/bundles.js +2 -0
- package/dist/src/domain/bundles.js.map +1 -0
- package/dist/src/domain/control-plane.js +2 -0
- package/dist/src/domain/control-plane.js.map +1 -0
- package/dist/src/domain/execution-evidence.js +2 -0
- package/dist/src/domain/execution-evidence.js.map +1 -0
- package/dist/src/domain/execution-plan.js +2 -0
- package/dist/src/domain/execution-plan.js.map +1 -0
- package/dist/src/domain/images.js +2 -0
- package/dist/src/domain/images.js.map +1 -0
- package/dist/src/domain/index.js +9 -0
- package/dist/src/domain/index.js.map +1 -1
- package/dist/src/domain/interactions.js +2 -0
- package/dist/src/domain/interactions.js.map +1 -0
- package/dist/src/domain/providers.js +2 -0
- package/dist/src/domain/providers.js.map +1 -0
- package/dist/src/domain/resources.js +2 -0
- package/dist/src/domain/resources.js.map +1 -0
- package/dist/src/domain/workers.js +2 -0
- package/dist/src/domain/workers.js.map +1 -0
- package/dist/src/evals/collect-only-rerun.js +182 -0
- package/dist/src/evals/collect-only-rerun.js.map +1 -0
- package/dist/src/evals/directory.js +30 -0
- package/dist/src/evals/directory.js.map +1 -0
- package/dist/src/evals/docker-engine-stats.js +89 -0
- package/dist/src/evals/docker-engine-stats.js.map +1 -0
- package/dist/src/evals/docker-ownership.js +51 -0
- package/dist/src/evals/docker-ownership.js.map +1 -0
- package/dist/src/evals/docker-reaper.js +144 -0
- package/dist/src/evals/docker-reaper.js.map +1 -0
- package/dist/src/evals/docker-resource-observer.js +207 -0
- package/dist/src/evals/docker-resource-observer.js.map +1 -0
- package/dist/src/evals/environment-image-planning.js +164 -0
- package/dist/src/evals/environment-image-planning.js.map +1 -0
- package/dist/src/evals/eval-artifact-planning.js +56 -0
- package/dist/src/evals/eval-artifact-planning.js.map +1 -0
- package/dist/src/evals/eval-finalization.js +10 -0
- package/dist/src/evals/eval-finalization.js.map +1 -0
- package/dist/src/evals/eval-lifecycle-events.js +16 -0
- package/dist/src/evals/eval-lifecycle-events.js.map +1 -0
- package/dist/src/evals/events.js +53 -5
- package/dist/src/evals/events.js.map +1 -1
- package/dist/src/evals/execution-evidence.js +116 -0
- package/dist/src/evals/execution-evidence.js.map +1 -0
- package/dist/src/evals/execution-leases.js +342 -0
- package/dist/src/evals/execution-leases.js.map +1 -0
- package/dist/src/evals/execution-plan-artifacts.js +72 -0
- package/dist/src/evals/execution-plan-artifacts.js.map +1 -0
- package/dist/src/evals/execution-plan-images.js +69 -0
- package/dist/src/evals/execution-plan-images.js.map +1 -0
- package/dist/src/evals/execution-plan-resources.js +174 -0
- package/dist/src/evals/execution-plan-resources.js.map +1 -0
- package/dist/src/evals/execution-plan.js +415 -0
- package/dist/src/evals/execution-plan.js.map +1 -0
- package/dist/src/evals/harbor-artifact-builder-image.js +99 -0
- package/dist/src/evals/harbor-artifact-builder-image.js.map +1 -0
- package/dist/src/evals/harbor-artifact-builder.js +354 -0
- package/dist/src/evals/harbor-artifact-builder.js.map +1 -0
- package/dist/src/evals/harbor-bridge-error.js +18 -3
- package/dist/src/evals/harbor-bridge-error.js.map +1 -1
- package/dist/src/evals/harbor-phase-timings.js +60 -0
- package/dist/src/evals/harbor-phase-timings.js.map +1 -0
- package/dist/src/evals/index.js +25 -2
- package/dist/src/evals/index.js.map +1 -1
- package/dist/src/evals/infrastructure-retry.js +100 -63
- package/dist/src/evals/infrastructure-retry.js.map +1 -1
- package/dist/src/evals/interaction-capture-import.js +49 -0
- package/dist/src/evals/interaction-capture-import.js.map +1 -0
- package/dist/src/evals/local-docker-provider.js +372 -0
- package/dist/src/evals/local-docker-provider.js.map +1 -0
- package/dist/src/evals/local-eval-planning.js +80 -0
- package/dist/src/evals/local-eval-planning.js.map +1 -0
- package/dist/src/evals/model-capture-plan.js +22 -0
- package/dist/src/evals/model-capture-plan.js.map +1 -0
- package/dist/src/evals/model-capture-runtime.js +114 -0
- package/dist/src/evals/model-capture-runtime.js.map +1 -0
- package/dist/src/evals/model-proxy-runtime-state.js +53 -0
- package/dist/src/evals/model-proxy-runtime-state.js.map +1 -0
- package/dist/src/evals/planned-execution-support.js +54 -0
- package/dist/src/evals/planned-execution-support.js.map +1 -0
- package/dist/src/evals/planned-execution.js +461 -0
- package/dist/src/evals/planned-execution.js.map +1 -0
- package/dist/src/evals/planned-remote-execution.js +40 -0
- package/dist/src/evals/planned-remote-execution.js.map +1 -0
- package/dist/src/evals/planned-retry-lifecycle.js +186 -0
- package/dist/src/evals/planned-retry-lifecycle.js.map +1 -0
- package/dist/src/evals/prepared-harness.js +76 -0
- package/dist/src/evals/prepared-harness.js.map +1 -0
- package/dist/src/evals/progress.js +4 -4
- package/dist/src/evals/progress.js.map +1 -1
- package/dist/src/evals/records.js +20 -3
- package/dist/src/evals/records.js.map +1 -1
- package/dist/src/evals/recovery.js +192 -0
- package/dist/src/evals/recovery.js.map +1 -0
- package/dist/src/evals/remote-infrastructure-retry.js +99 -0
- package/dist/src/evals/remote-infrastructure-retry.js.map +1 -0
- package/dist/src/evals/rerun-inputs.js +89 -0
- package/dist/src/evals/rerun-inputs.js.map +1 -0
- package/dist/src/evals/rerun-slots.js +4 -3
- package/dist/src/evals/rerun-slots.js.map +1 -1
- package/dist/src/evals/rerun-types.js +49 -0
- package/dist/src/evals/rerun-types.js.map +1 -0
- package/dist/src/evals/rerun.js +113 -75
- package/dist/src/evals/rerun.js.map +1 -1
- package/dist/src/evals/result-helpers.js +65 -0
- package/dist/src/evals/result-helpers.js.map +1 -0
- package/dist/src/evals/resume-state.js +36 -0
- package/dist/src/evals/resume-state.js.map +1 -0
- package/dist/src/evals/service-types.js +2 -0
- package/dist/src/evals/service-types.js.map +1 -0
- package/dist/src/evals/service.js +292 -257
- package/dist/src/evals/service.js.map +1 -1
- package/dist/src/evals/task-resources.js +209 -0
- package/dist/src/evals/task-resources.js.map +1 -0
- package/dist/src/evals/trial-environment-evidence.js +87 -0
- package/dist/src/evals/trial-environment-evidence.js.map +1 -0
- package/dist/src/evals/trial-execution-evidence.js +13 -0
- package/dist/src/evals/trial-execution-evidence.js.map +1 -0
- package/dist/src/evals/trial-import-identity.js +31 -0
- package/dist/src/evals/trial-import-identity.js.map +1 -0
- package/dist/src/evals/trial-import.js +143 -145
- package/dist/src/evals/trial-import.js.map +1 -1
- package/dist/src/evals/trial-publication-recovery.js +104 -0
- package/dist/src/evals/trial-publication-recovery.js.map +1 -0
- package/dist/src/evals/trial-publication.js +25 -0
- package/dist/src/evals/trial-publication.js.map +1 -0
- package/dist/src/evals/work-item-artifacts.js +23 -0
- package/dist/src/evals/work-item-artifacts.js.map +1 -0
- package/dist/src/foundation/config.js +17 -0
- package/dist/src/foundation/config.js.map +1 -1
- package/dist/src/foundation/credential-redaction.js +137 -0
- package/dist/src/foundation/credential-redaction.js.map +1 -0
- package/dist/src/foundation/environment-image-references.js +59 -0
- package/dist/src/foundation/environment-image-references.js.map +1 -0
- package/dist/src/foundation/executable.js +45 -5
- package/dist/src/foundation/executable.js.map +1 -1
- package/dist/src/foundation/index.js +5 -2
- package/dist/src/foundation/index.js.map +1 -1
- package/dist/src/foundation/process-identity.js +69 -0
- package/dist/src/foundation/process-identity.js.map +1 -0
- package/dist/src/images/context.js +47 -0
- package/dist/src/images/context.js.map +1 -0
- package/dist/src/images/docker-buildkit.js +137 -0
- package/dist/src/images/docker-buildkit.js.map +1 -0
- package/dist/src/images/dockerfile.js +67 -0
- package/dist/src/images/dockerfile.js.map +1 -0
- package/dist/src/images/gc.js +261 -0
- package/dist/src/images/gc.js.map +1 -0
- package/dist/src/images/index.js +10 -0
- package/dist/src/images/index.js.map +1 -0
- package/dist/src/images/manifest.js +109 -0
- package/dist/src/images/manifest.js.map +1 -0
- package/dist/src/images/ownership.js +5 -0
- package/dist/src/images/ownership.js.map +1 -0
- package/dist/src/images/records.js +32 -0
- package/dist/src/images/records.js.map +1 -0
- package/dist/src/images/registry.js +154 -0
- package/dist/src/images/registry.js.map +1 -0
- package/dist/src/images/service.js +224 -0
- package/dist/src/images/service.js.map +1 -0
- package/dist/src/model-access/capture.js +255 -0
- package/dist/src/model-access/capture.js.map +1 -0
- package/dist/src/model-access/index.js +5 -0
- package/dist/src/model-access/index.js.map +1 -0
- package/dist/src/model-access/policy.js +49 -0
- package/dist/src/model-access/policy.js.map +1 -0
- package/dist/src/model-access/proxy.js +336 -0
- package/dist/src/model-access/proxy.js.map +1 -0
- package/dist/src/model-access/records.js +167 -0
- package/dist/src/model-access/records.js.map +1 -0
- package/dist/src/runs/bundle.js +450 -0
- package/dist/src/runs/bundle.js.map +1 -0
- package/dist/src/runs/events.js +12 -1
- package/dist/src/runs/events.js.map +1 -1
- package/dist/src/runs/executor.js +24 -17
- package/dist/src/runs/executor.js.map +1 -1
- package/dist/src/runs/index.js +3 -1
- package/dist/src/runs/index.js.map +1 -1
- package/dist/src/runs/manifest.js +5 -5
- package/dist/src/runs/manifest.js.map +1 -1
- package/dist/src/runs/request.js +6 -1
- package/dist/src/runs/request.js.map +1 -1
- package/dist/src/runs/training-candidate.js +161 -0
- package/dist/src/runs/training-candidate.js.map +1 -0
- package/dist/src/trajectories/provider-capture.js +15 -24
- package/dist/src/trajectories/provider-capture.js.map +1 -1
- package/dist/src/trajectories/providers/deepseek.js +3 -3
- package/dist/src/trajectories/providers/deepseek.js.map +1 -1
- package/dist/src/workers/index.js +3 -0
- package/dist/src/workers/index.js.map +1 -0
- package/dist/src/workers/remote-harbor-work-spec.js +97 -0
- package/dist/src/workers/remote-harbor-work-spec.js.map +1 -0
- package/dist/src/workers/remote-harbor-worker.js +220 -0
- package/dist/src/workers/remote-harbor-worker.js.map +1 -0
- package/docs/schemas/environment-build-index.schema.json +13 -0
- package/docs/schemas/environment-build-record.schema.json +25 -0
- package/docs/schemas/environment-image-manifest.schema.json +62 -0
- package/docs/schemas/eval-control.schema.json +36 -0
- package/docs/schemas/eval-execution-plan.schema.json +216 -0
- package/docs/schemas/eval-rerun-result.schema.json +53 -0
- package/docs/schemas/eval-rerun-state.schema.json +61 -0
- package/docs/schemas/eval-rerun-submission.schema.json +58 -0
- package/docs/schemas/eval-result.schema.json +34 -1
- package/docs/schemas/eval-submission-input.schema.json +14 -0
- package/docs/schemas/eval-submission.schema.json +68 -0
- package/docs/schemas/eval-trial-publication.schema.json +39 -0
- package/docs/schemas/execution-evidence.schema.json +96 -0
- package/docs/schemas/execution-lease.schema.json +50 -0
- package/docs/schemas/execution-provider-status.schema.json +65 -0
- package/docs/schemas/execution-worker.schema.json +53 -0
- package/docs/schemas/harbor-process-exit.schema.json +14 -0
- package/docs/schemas/interaction-capture-ref.schema.json +45 -0
- package/docs/schemas/local-provider-execution.schema.json +33 -0
- package/docs/schemas/model-interaction.schema.json +41 -0
- package/docs/schemas/remote-result-envelope.schema.json +32 -0
- package/docs/schemas/remote-tree-envelope.schema.json +25 -0
- package/docs/schemas/remote-work-offer.schema.json +58 -0
- package/docs/schemas/remote-work-receipt.schema.json +50 -0
- package/docs/schemas/remote-worker-artifact.schema.json +17 -0
- package/docs/schemas/remote-worker-event.schema.json +18 -0
- package/docs/schemas/remote-worker-heartbeat.schema.json +37 -0
- package/docs/schemas/remote-worker-registration.schema.json +52 -0
- package/docs/schemas/result-bundle-index.schema.json +133 -0
- package/docs/schemas/run-environment-images.schema.json +39 -0
- package/docs/schemas/training-data-candidate.schema.json +46 -0
- package/integrations/harbor/hitch_harbor_agent.py +171 -539
- package/integrations/harbor/hitch_harbor_environment.py +331 -0
- package/integrations/harbor/hitch_harbor_task_resources.py +275 -0
- package/integrations/harbor/hitch_harbor_verifier.py +23 -0
- package/package.json +14 -1
|
@@ -0,0 +1,40 @@
|
|
|
1
|
+
import { preparedArtifactForWorkItem } from "./work-item-artifacts.js";
|
|
2
|
+
import { loadTrialEnvironmentImages } from "./trial-environment-evidence.js";
|
|
3
|
+
export async function executeRemotePlannedWorkItem(input) {
|
|
4
|
+
const { options, item } = input;
|
|
5
|
+
if (!options.remoteWorkExecutor)
|
|
6
|
+
throw new TypeError("remote work executor is unavailable");
|
|
7
|
+
const taskId = item.task_ids[0];
|
|
8
|
+
const environmentImages = await loadTrialEnvironmentImages({
|
|
9
|
+
taskId, uses: item.image_refs ?? [],
|
|
10
|
+
...(options.environmentImageManifestLoader ? { loader: options.environmentImageManifestLoader } : {}),
|
|
11
|
+
});
|
|
12
|
+
const completed = await options.remoteWorkExecutor({
|
|
13
|
+
evalId: options.evalId,
|
|
14
|
+
evalDirectory: options.evalDirectory,
|
|
15
|
+
root: options.root,
|
|
16
|
+
request: options.request,
|
|
17
|
+
plan: options.plan,
|
|
18
|
+
workItem: item,
|
|
19
|
+
resolvedRevision: options.resolvedRevision,
|
|
20
|
+
preparedArtifact: preparedArtifactForWorkItem(options, item),
|
|
21
|
+
runtimeDirectory: options.controllerRuntime.directory,
|
|
22
|
+
runtimeId: options.controllerRuntime.runtime_id,
|
|
23
|
+
...(environmentImages ? { environmentImages } : {}),
|
|
24
|
+
...(options.plan.model_capture ? { modelCapturePlan: options.plan.model_capture } : {}),
|
|
25
|
+
...(options.signal ? { signal: options.signal } : {}),
|
|
26
|
+
emit: (event) => options.sink.emit(event),
|
|
27
|
+
publish: input.publish,
|
|
28
|
+
onLeaseState: (leaseId, state) => options.onWorkItemState?.(item.work_id, leaseId, state) ?? Promise.resolve(),
|
|
29
|
+
});
|
|
30
|
+
return {
|
|
31
|
+
attempt: item.logical_attempt,
|
|
32
|
+
workId: item.work_id,
|
|
33
|
+
tasks: [taskId],
|
|
34
|
+
refs: completed.refs,
|
|
35
|
+
leaseId: completed.leaseId,
|
|
36
|
+
run: completed.run,
|
|
37
|
+
...(environmentImages ? { environmentImages } : {}),
|
|
38
|
+
};
|
|
39
|
+
}
|
|
40
|
+
//# sourceMappingURL=planned-remote-execution.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"planned-remote-execution.js","sourceRoot":"","sources":["../../../src/evals/planned-remote-execution.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,2BAA2B,EAAE,MAAM,0BAA0B,CAAC;AACvE,OAAO,EAAE,0BAA0B,EAAE,MAAM,iCAAiC,CAAC;AAE7E,MAAM,CAAC,KAAK,UAAU,4BAA4B,CAAC,KAIlD;IACC,MAAM,EAAE,OAAO,EAAE,IAAI,EAAE,GAAG,KAAK,CAAC;IAChC,IAAI,CAAC,OAAO,CAAC,kBAAkB;QAAE,MAAM,IAAI,SAAS,CAAC,qCAAqC,CAAC,CAAC;IAC5F,MAAM,MAAM,GAAG,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAW,CAAC;IAC1C,MAAM,iBAAiB,GAAG,MAAM,0BAA0B,CAAC;QACzD,MAAM,EAAE,IAAI,EAAE,IAAI,CAAC,UAAU,IAAI,EAAE;QACnC,GAAG,CAAC,OAAO,CAAC,8BAA8B,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,8BAA8B,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KACtG,CAAC,CAAC;IACH,MAAM,SAAS,GAAG,MAAM,OAAO,CAAC,kBAAkB,CAAC;QACjD,MAAM,EAAE,OAAO,CAAC,MAAM;QACtB,aAAa,EAAE,OAAO,CAAC,aAAa;QACpC,IAAI,EAAE,OAAO,CAAC,IAAI;QAClB,OAAO,EAAE,OAAO,CAAC,OAAO;QACxB,IAAI,EAAE,OAAO,CAAC,IAAI;QAClB,QAAQ,EAAE,IAAI;QACd,gBAAgB,EAAE,OAAO,CAAC,gBAAgB;QAC1C,gBAAgB,EAAE,2BAA2B,CAAC,OAAO,EAAE,IAAI,CAAC;QAC5D,gBAAgB,EAAE,OAAO,CAAC,iBAAiB,CAAC,SAAS;QACrD,SAAS,EAAE,OAAO,CAAC,iBAAiB,CAAC,UAAU;QAC/C,GAAG,CAAC,iBAAiB,CAAC,CAAC,CAAC,EAAE,iBAAiB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACnD,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,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACrD,IAAI,EAAE,CAAC,KAAK,EAAE,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC;QACzC,OAAO,EAAE,KAAK,CAAC,OAAO;QACtB,YAAY,EAAE,CAAC,OAAO,EAAE,KAAK,EAAE,EAAE,CAAC,OAAO,CAAC,eAAe,EAAE,CAAC,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,KAAK,CAAC,IAAI,OAAO,CAAC,OAAO,EAAE;KAC/G,CAAC,CAAC;IACH,OAAO;QACL,OAAO,EAAE,IAAI,CAAC,eAAyB;QACvC,MAAM,EAAE,IAAI,CAAC,OAAO;QACpB,KAAK,EAAE,CAAC,MAAM,CAAC;QACf,IAAI,EAAE,SAAS,CAAC,IAAI;QACpB,OAAO,EAAE,SAAS,CAAC,OAAO;QAC1B,GAAG,EAAE,SAAS,CAAC,GAAG;QAClB,GAAG,CAAC,iBAAiB,CAAC,CAAC,CAAC,EAAE,iBAAiB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KACpD,CAAC;AACJ,CAAC"}
|
|
@@ -0,0 +1,186 @@
|
|
|
1
|
+
import { dockerResourceOwnership } from "./docker-ownership.js";
|
|
2
|
+
import { startDockerResourceObserver } from "./docker-resource-observer.js";
|
|
3
|
+
import { DEFAULT_EXECUTION_LEASE_HEARTBEAT_MS, DEFAULT_EXECUTION_LEASE_TTL_MS, createExecutionLease } from "./execution-leases.js";
|
|
4
|
+
import { resourceRequirementForTask, runtimeResourcesForTask } from "./execution-plan-resources.js";
|
|
5
|
+
import { recordLocalDockerProcessExit, recordLocalDockerProcessStart, releaseLocalDockerProcessRecord } from "./local-docker-provider.js";
|
|
6
|
+
import { resolvedImageMapping } from "./environment-image-planning.js";
|
|
7
|
+
import { prebuiltTaskImage, verifyTrialEnvironmentImageExecution } from "./trial-environment-evidence.js";
|
|
8
|
+
export async function beginPlannedInfrastructureRetry(input) {
|
|
9
|
+
const { options, item } = input;
|
|
10
|
+
const taskId = item.task_ids[0];
|
|
11
|
+
let permit;
|
|
12
|
+
let lease;
|
|
13
|
+
let runningAnnounced = false;
|
|
14
|
+
try {
|
|
15
|
+
permit = await options.admission?.acquire({
|
|
16
|
+
evalId: options.evalId,
|
|
17
|
+
workItem: item,
|
|
18
|
+
maxParallelism: options.plan.max_parallelism,
|
|
19
|
+
...(options.signal ? { signal: options.signal } : {}),
|
|
20
|
+
});
|
|
21
|
+
if (permit)
|
|
22
|
+
options.sink.emit({
|
|
23
|
+
type: "eval.physical-retry.admitted",
|
|
24
|
+
execution_kind: "physical-infrastructure-retry",
|
|
25
|
+
work_id: item.work_id,
|
|
26
|
+
retry: input.retry,
|
|
27
|
+
allocation_id: permit.allocationId,
|
|
28
|
+
collision_keys: permit.collisionKeys,
|
|
29
|
+
reservation: item.reservation,
|
|
30
|
+
});
|
|
31
|
+
lease = await createExecutionLease({
|
|
32
|
+
evalDirectory: options.evalDirectory,
|
|
33
|
+
evalId: options.evalId,
|
|
34
|
+
workId: item.work_id,
|
|
35
|
+
worker: permit ? { ...options.worker, parentAllocationId: permit.allocationId } : options.worker,
|
|
36
|
+
reservation: item.reservation,
|
|
37
|
+
ttlMs: DEFAULT_EXECUTION_LEASE_TTL_MS,
|
|
38
|
+
});
|
|
39
|
+
const epoch = lease.current().epoch;
|
|
40
|
+
await lease.markRunning(epoch);
|
|
41
|
+
await options.onWorkItemState?.(item.work_id, lease.leaseId, "running");
|
|
42
|
+
runningAnnounced = true;
|
|
43
|
+
options.sink.emit({
|
|
44
|
+
type: "eval.physical-retry.started",
|
|
45
|
+
execution_kind: "physical-infrastructure-retry",
|
|
46
|
+
candidate_executes: true,
|
|
47
|
+
work_id: item.work_id,
|
|
48
|
+
lease_id: lease.leaseId,
|
|
49
|
+
lease_epoch: epoch,
|
|
50
|
+
retry: input.retry,
|
|
51
|
+
task_id: taskId,
|
|
52
|
+
attempt: item.logical_attempt,
|
|
53
|
+
trigger_trials: input.triggers.map((trial) => trial.trial_id).sort(),
|
|
54
|
+
});
|
|
55
|
+
return retryLifecycle(input, lease, permit, runningAnnounced);
|
|
56
|
+
}
|
|
57
|
+
catch (error) {
|
|
58
|
+
if (lease)
|
|
59
|
+
await lease.release(lease.current().epoch).catch(() => undefined);
|
|
60
|
+
if (runningAnnounced)
|
|
61
|
+
await options.onWorkItemState?.(item.work_id, lease?.leaseId ?? "", "terminal").catch(() => undefined);
|
|
62
|
+
permit?.release();
|
|
63
|
+
throw error;
|
|
64
|
+
}
|
|
65
|
+
}
|
|
66
|
+
function retryLifecycle(input, lease, permit, runningAnnounced) {
|
|
67
|
+
const { options, item } = input;
|
|
68
|
+
const taskId = item.task_ids[0];
|
|
69
|
+
const epoch = lease.current().epoch;
|
|
70
|
+
const runtime = runtimeResourcesForTask(options.plan, taskId, item.reservation);
|
|
71
|
+
const ownership = dockerResourceOwnership(options.root, lease.current(), taskId);
|
|
72
|
+
const observer = startDockerResourceObserver({
|
|
73
|
+
ownership,
|
|
74
|
+
workerId: options.worker.workerId,
|
|
75
|
+
collisionDomainId: options.worker.collisionDomainId,
|
|
76
|
+
reservation: item.reservation,
|
|
77
|
+
mainLimits: runtime.mainLimits,
|
|
78
|
+
sidecarLimits: runtime.sidecarLimits,
|
|
79
|
+
env: options.env,
|
|
80
|
+
...(options.signal ? { signal: options.signal } : {}),
|
|
81
|
+
});
|
|
82
|
+
let processRecorded = false;
|
|
83
|
+
let heartbeatFailure;
|
|
84
|
+
let heartbeatTail = Promise.resolve();
|
|
85
|
+
const timer = setInterval(() => {
|
|
86
|
+
if (heartbeatFailure !== undefined)
|
|
87
|
+
return;
|
|
88
|
+
heartbeatTail = heartbeatTail.then(async () => {
|
|
89
|
+
const renewed = await lease.heartbeat(epoch);
|
|
90
|
+
options.sink.emit({
|
|
91
|
+
type: "lease.renewed",
|
|
92
|
+
execution_kind: "physical-infrastructure-retry",
|
|
93
|
+
work_id: item.work_id,
|
|
94
|
+
lease_id: lease.leaseId,
|
|
95
|
+
lease_epoch: renewed.epoch,
|
|
96
|
+
heartbeat_at: renewed.heartbeat_at,
|
|
97
|
+
expires_at: renewed.expires_at,
|
|
98
|
+
});
|
|
99
|
+
}).catch((error) => { heartbeatFailure ??= error; });
|
|
100
|
+
}, DEFAULT_EXECUTION_LEASE_HEARTBEAT_MS);
|
|
101
|
+
timer.unref();
|
|
102
|
+
let closed;
|
|
103
|
+
const prebuiltImage = prebuiltTaskImage(input.environmentImages);
|
|
104
|
+
const close = () => closed ??= closeRetryLifecycle({ input, lease, permit, observer, timer, heartbeatTail: () => heartbeatTail, heartbeatFailure: () => heartbeatFailure, processRecorded: () => processRecorded, runningAnnounced });
|
|
105
|
+
return {
|
|
106
|
+
leaseId: lease.leaseId,
|
|
107
|
+
workId: item.work_id,
|
|
108
|
+
...(input.environmentImages ? { environmentImages: input.environmentImages } : {}),
|
|
109
|
+
captureExecutionEvidence: async () => {
|
|
110
|
+
if (heartbeatFailure !== undefined)
|
|
111
|
+
throw heartbeatFailure;
|
|
112
|
+
return verifyTrialEnvironmentImageExecution(await observer.capture(), input.environmentImages);
|
|
113
|
+
},
|
|
114
|
+
backend: {
|
|
115
|
+
executionResources: resourceRequirementForTask(options.plan, taskId)?.main_limits ?? options.plan.default_trial_resources,
|
|
116
|
+
...(Object.keys(runtime.sidecarLimits).length > 0 ? { dockerServiceLimits: runtime.sidecarLimits } : {}),
|
|
117
|
+
dockerOwnership: ownership,
|
|
118
|
+
resolvedImages: resolvedImageMapping(item.image_refs ?? []),
|
|
119
|
+
...(prebuiltImage ? { prebuiltTaskImage: prebuiltImage } : {}),
|
|
120
|
+
...(options.worker.provider === "local-docker" && process.platform !== "win32" ? {
|
|
121
|
+
recoverableProcess: true,
|
|
122
|
+
onProcessStarted: (pid) => recordLocalDockerProcessStart({
|
|
123
|
+
root: options.root,
|
|
124
|
+
workerId: options.worker.workerId,
|
|
125
|
+
evalDirectory: options.evalDirectory,
|
|
126
|
+
lease: lease.current(),
|
|
127
|
+
backendDirectory: input.backendDirectory,
|
|
128
|
+
pid,
|
|
129
|
+
}).then(() => { processRecorded = true; }),
|
|
130
|
+
onProcessExited: (result) => recordLocalDockerProcessExit({
|
|
131
|
+
root: options.root,
|
|
132
|
+
leaseId: lease.leaseId,
|
|
133
|
+
epoch,
|
|
134
|
+
...result,
|
|
135
|
+
}).then(() => undefined),
|
|
136
|
+
} : {}),
|
|
137
|
+
},
|
|
138
|
+
close,
|
|
139
|
+
};
|
|
140
|
+
}
|
|
141
|
+
async function closeRetryLifecycle(input) {
|
|
142
|
+
const { options, item } = input.input;
|
|
143
|
+
const epoch = input.lease.current().epoch;
|
|
144
|
+
let failure;
|
|
145
|
+
const attempt = async (operation) => {
|
|
146
|
+
try {
|
|
147
|
+
await operation();
|
|
148
|
+
}
|
|
149
|
+
catch (error) {
|
|
150
|
+
failure ??= error;
|
|
151
|
+
}
|
|
152
|
+
};
|
|
153
|
+
clearInterval(input.timer);
|
|
154
|
+
await attempt(() => input.observer.stop());
|
|
155
|
+
await attempt(input.heartbeatTail);
|
|
156
|
+
failure ??= input.heartbeatFailure();
|
|
157
|
+
if (input.processRecorded())
|
|
158
|
+
await attempt(() => releaseLocalDockerProcessRecord({ root: options.root, leaseId: input.lease.leaseId, epoch }));
|
|
159
|
+
await attempt(async () => {
|
|
160
|
+
const released = await input.lease.release(epoch);
|
|
161
|
+
options.sink.emit({
|
|
162
|
+
type: "eval.physical-retry.lease-released",
|
|
163
|
+
execution_kind: "physical-infrastructure-retry",
|
|
164
|
+
work_id: item.work_id,
|
|
165
|
+
lease_id: input.lease.leaseId,
|
|
166
|
+
lease_epoch: released.epoch,
|
|
167
|
+
retry: input.input.retry,
|
|
168
|
+
state: released.state,
|
|
169
|
+
});
|
|
170
|
+
});
|
|
171
|
+
if (options.dockerResourceReaper) {
|
|
172
|
+
try {
|
|
173
|
+
const report = await options.dockerResourceReaper({ root: options.root, leaseIds: [input.lease.leaseId], env: options.env });
|
|
174
|
+
options.sink.emit({ type: "eval.physical-retry.resources-reaped", work_id: item.work_id, lease_id: input.lease.leaseId, retry: input.input.retry, scanned: report.scanned, deleted: report.deleted.length, issues: report.issues.length });
|
|
175
|
+
}
|
|
176
|
+
catch (error) {
|
|
177
|
+
options.sink.emit({ type: "eval.physical-retry.reaper-failed", work_id: item.work_id, lease_id: input.lease.leaseId, retry: input.input.retry, code: error.code || "docker_reaper_failed" });
|
|
178
|
+
}
|
|
179
|
+
}
|
|
180
|
+
if (input.runningAnnounced)
|
|
181
|
+
await attempt(() => options.onWorkItemState?.(item.work_id, input.lease.leaseId, "terminal") ?? Promise.resolve());
|
|
182
|
+
input.permit?.release();
|
|
183
|
+
if (failure !== undefined)
|
|
184
|
+
throw failure;
|
|
185
|
+
}
|
|
186
|
+
//# sourceMappingURL=planned-retry-lifecycle.js.map
|
|
@@ -0,0 +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;AAG1G,MAAM,CAAC,KAAK,UAAU,+BAA+B,CAAC,KAOrD;IACC,MAAM,EAAE,OAAO,EAAE,IAAI,EAAE,GAAG,KAAK,CAAC;IAChC,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,KAAK,EAAE,KAAK,EAAE,MAAM,EAAE,gBAAgB,CAAC,CAAC;IAChE,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 +1,76 @@
|
|
|
1
|
+
import { preparedArtifactDirectory } from "../artifacts/index.js";
|
|
2
|
+
import { HitchError } from "../foundation/index.js";
|
|
3
|
+
import { prepareHarness } from "../artifacts/index.js";
|
|
4
|
+
export function harborPreparedArtifact(root, artifact) {
|
|
5
|
+
if (!artifact.artifact_integrity || !artifact.entrypoint_integrity) {
|
|
6
|
+
throw new HitchError("host-prepared harness artifact has no complete integrity metadata", {
|
|
7
|
+
code: "artifact_integrity_mismatch",
|
|
8
|
+
exitCode: 5,
|
|
9
|
+
});
|
|
10
|
+
}
|
|
11
|
+
return {
|
|
12
|
+
directory: preparedArtifactDirectory(root, artifact.artifact_id),
|
|
13
|
+
artifact_id: artifact.artifact_id,
|
|
14
|
+
artifact_integrity: artifact.artifact_integrity,
|
|
15
|
+
entrypoint_integrity: artifact.entrypoint_integrity,
|
|
16
|
+
harness_id: artifact.harness_id,
|
|
17
|
+
revision_identity: artifact.revision_identity,
|
|
18
|
+
adapter_version: artifact.adapter_version,
|
|
19
|
+
recipe_version: artifact.recipe_version,
|
|
20
|
+
platform: artifact.platform,
|
|
21
|
+
node_version: artifact.toolchain.node || process.version,
|
|
22
|
+
source_type: artifact.source_type,
|
|
23
|
+
storage: "host-artifact-store-v1",
|
|
24
|
+
};
|
|
25
|
+
}
|
|
26
|
+
export function preparedHarnessEvent(artifact, cacheHit, source, builder) {
|
|
27
|
+
return {
|
|
28
|
+
type: "eval.harness-artifact.prepared",
|
|
29
|
+
harness: artifact.harness_id,
|
|
30
|
+
artifact_id: artifact.artifact_id,
|
|
31
|
+
platform: artifact.platform,
|
|
32
|
+
cache_hit: cacheHit,
|
|
33
|
+
source,
|
|
34
|
+
...(builder?.image ? { builder_image: builder.image } : {}),
|
|
35
|
+
...(builder?.imageId ? { builder_image_id: builder.imageId } : {}),
|
|
36
|
+
};
|
|
37
|
+
}
|
|
38
|
+
export async function prepareEvalHarborArtifact(input) {
|
|
39
|
+
const selector = input.requestedReference.selector;
|
|
40
|
+
const verifiedLocalGitSource = input.localTransport && selector.type === "commit" && selector.source?.explicit
|
|
41
|
+
? {
|
|
42
|
+
directory: selector.source.local_path,
|
|
43
|
+
commit: input.localTransport.manifest.commit,
|
|
44
|
+
tree: input.localTransport.manifest.tree,
|
|
45
|
+
resolutionIdentity: input.localTransport.manifest.resolution_identity,
|
|
46
|
+
payloadSha256: input.localTransport.manifest.payload_sha256,
|
|
47
|
+
}
|
|
48
|
+
: undefined;
|
|
49
|
+
return input.builder({
|
|
50
|
+
root: input.root,
|
|
51
|
+
resolvedRevision: input.resolvedRevision,
|
|
52
|
+
runtimeDirectory: input.controllerRuntime.directory,
|
|
53
|
+
runtimeId: input.controllerRuntime.runtime_id,
|
|
54
|
+
runtimeContract: input.runtimeContract,
|
|
55
|
+
...(input.localTransport ? { localTransport: input.localTransport } : {}),
|
|
56
|
+
...(verifiedLocalGitSource ? { verifiedLocalGitSource } : {}),
|
|
57
|
+
env: input.env,
|
|
58
|
+
...(input.signal ? { signal: input.signal } : {}),
|
|
59
|
+
});
|
|
60
|
+
}
|
|
61
|
+
/** In-process seam for tests whose fake package manager cannot be mounted into
|
|
62
|
+
* Docker. Production callers must use the dedicated builder. */
|
|
63
|
+
export const prepareHostHarborArtifactForTest = async (input) => {
|
|
64
|
+
const artifact = await prepareHarness(input.resolvedRevision, {
|
|
65
|
+
root: input.root,
|
|
66
|
+
env: input.env,
|
|
67
|
+
...(input.signal ? { signal: input.signal } : {}),
|
|
68
|
+
...(input.verifiedLocalGitSource ? { verifiedLocalGitSource: input.verifiedLocalGitSource } : {}),
|
|
69
|
+
});
|
|
70
|
+
return {
|
|
71
|
+
artifact: harborPreparedArtifact(input.root, artifact),
|
|
72
|
+
cacheHit: artifact.cache_hit,
|
|
73
|
+
source: "test-host",
|
|
74
|
+
};
|
|
75
|
+
};
|
|
76
|
+
//# sourceMappingURL=prepared-harness.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"prepared-harness.js","sourceRoot":"","sources":["../../../src/evals/prepared-harness.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,yBAAyB,EAAE,MAAM,uBAAuB,CAAC;AAIlE,OAAO,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAEpD,OAAO,EAAE,cAAc,EAAE,MAAM,uBAAuB,CAAC;AAKvD,MAAM,UAAU,sBAAsB,CAAC,IAAY,EAAE,QAA0B;IAC7E,IAAI,CAAC,QAAQ,CAAC,kBAAkB,IAAI,CAAC,QAAQ,CAAC,oBAAoB,EAAE,CAAC;QACnE,MAAM,IAAI,UAAU,CAAC,mEAAmE,EAAE;YACxF,IAAI,EAAE,6BAA6B;YACnC,QAAQ,EAAE,CAAC;SACZ,CAAC,CAAC;IACL,CAAC;IACD,OAAO;QACL,SAAS,EAAE,yBAAyB,CAAC,IAAI,EAAE,QAAQ,CAAC,WAAW,CAAC;QAChE,WAAW,EAAE,QAAQ,CAAC,WAAW;QACjC,kBAAkB,EAAE,QAAQ,CAAC,kBAAkB;QAC/C,oBAAoB,EAAE,QAAQ,CAAC,oBAAoB;QACnD,UAAU,EAAE,QAAQ,CAAC,UAAU;QAC/B,iBAAiB,EAAE,QAAQ,CAAC,iBAAiB;QAC7C,eAAe,EAAE,QAAQ,CAAC,eAAe;QACzC,cAAc,EAAE,QAAQ,CAAC,cAAc;QACvC,QAAQ,EAAE,QAAQ,CAAC,QAAQ;QAC3B,YAAY,EAAE,QAAQ,CAAC,SAAS,CAAC,IAAI,IAAI,OAAO,CAAC,OAAO;QACxD,WAAW,EAAE,QAAQ,CAAC,WAAW;QACjC,OAAO,EAAE,wBAAwB;KAClC,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,oBAAoB,CAClC,QAA2E,EAC3E,QAAiB,EACjB,MAAyC,EACzC,OAA8C;IAE9C,OAAO;QACL,IAAI,EAAE,gCAAgC;QACtC,OAAO,EAAE,QAAQ,CAAC,UAAU;QAC5B,WAAW,EAAE,QAAQ,CAAC,WAAW;QACjC,QAAQ,EAAE,QAAQ,CAAC,QAAQ;QAC3B,SAAS,EAAE,QAAQ;QACnB,MAAM;QACN,GAAG,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC,EAAE,aAAa,EAAE,OAAO,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC3D,GAAG,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC,CAAC,EAAE,gBAAgB,EAAE,OAAO,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KACnE,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAAC,KAU/C;IACC,MAAM,QAAQ,GAAG,KAAK,CAAC,kBAAkB,CAAC,QAAQ,CAAC;IACnD,MAAM,sBAAsB,GAAG,KAAK,CAAC,cAAc,IAAI,QAAQ,CAAC,IAAI,KAAK,QAAQ,IAAI,QAAQ,CAAC,MAAM,EAAE,QAAQ;QAC5G,CAAC,CAAC;YACE,SAAS,EAAE,QAAQ,CAAC,MAAM,CAAC,UAAU;YACrC,MAAM,EAAE,KAAK,CAAC,cAAc,CAAC,QAAQ,CAAC,MAAM;YAC5C,IAAI,EAAE,KAAK,CAAC,cAAc,CAAC,QAAQ,CAAC,IAAI;YACxC,kBAAkB,EAAE,KAAK,CAAC,cAAc,CAAC,QAAQ,CAAC,mBAAmB;YACrE,aAAa,EAAE,KAAK,CAAC,cAAc,CAAC,QAAQ,CAAC,cAAc;SAC5D;QACH,CAAC,CAAC,SAAS,CAAC;IACd,OAAO,KAAK,CAAC,OAAO,CAAC;QACnB,IAAI,EAAE,KAAK,CAAC,IAAI;QAChB,gBAAgB,EAAE,KAAK,CAAC,gBAAgB;QACxC,gBAAgB,EAAE,KAAK,CAAC,iBAAiB,CAAC,SAAS;QACnD,SAAS,EAAE,KAAK,CAAC,iBAAiB,CAAC,UAAU;QAC7C,eAAe,EAAE,KAAK,CAAC,eAAe;QACtC,GAAG,CAAC,KAAK,CAAC,cAAc,CAAC,CAAC,CAAC,EAAE,cAAc,EAAE,KAAK,CAAC,cAAc,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACzE,GAAG,CAAC,sBAAsB,CAAC,CAAC,CAAC,EAAE,sBAAsB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC7D,GAAG,EAAE,KAAK,CAAC,GAAG;QACd,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAClD,CAAC,CAAC;AACL,CAAC;AAED;gEACgE;AAChE,MAAM,CAAC,MAAM,gCAAgC,GAA8B,KAAK,EAAE,KAAK,EAAE,EAAE;IACzF,MAAM,QAAQ,GAAG,MAAM,cAAc,CAAC,KAAK,CAAC,gBAAgB,EAAE;QAC5D,IAAI,EAAE,KAAK,CAAC,IAAI;QAChB,GAAG,EAAE,KAAK,CAAC,GAAG;QACd,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACjD,GAAG,CAAC,KAAK,CAAC,sBAAsB,CAAC,CAAC,CAAC,EAAE,sBAAsB,EAAE,KAAK,CAAC,sBAAsB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAClG,CAAC,CAAC;IACH,OAAO;QACL,QAAQ,EAAE,sBAAsB,CAAC,KAAK,CAAC,IAAI,EAAE,QAAQ,CAAC;QACtD,QAAQ,EAAE,QAAQ,CAAC,SAAS;QAC5B,MAAM,EAAE,WAAW;KACpB,CAAC;AACJ,CAAC,CAAC"}
|
|
@@ -32,7 +32,7 @@ export async function writeEvalProgress(evalDirectory, progress) {
|
|
|
32
32
|
await atomicWriteJSON(path.join(evalDirectory, "progress.json"), parseEvalProgress(progress));
|
|
33
33
|
}
|
|
34
34
|
export function mergeEvalProgressTrial(progress, trial, now = new Date().toISOString()) {
|
|
35
|
-
const parsed =
|
|
35
|
+
const parsed = parseEvalTrialRef(trial, "eval progress trial");
|
|
36
36
|
const byTrial = progress.trials.find((item) => item.trial_id === parsed.trial_id);
|
|
37
37
|
if (byTrial !== undefined) {
|
|
38
38
|
if (JSON.stringify(byTrial) !== JSON.stringify(parsed))
|
|
@@ -66,7 +66,7 @@ export function evalTrialKey(trial) {
|
|
|
66
66
|
}
|
|
67
67
|
/** Replace one invalid logical trial, or fill a missing slot, with a valid verifier result. */
|
|
68
68
|
export function replaceInvalidEvalProgressTrial(progress, trial, now = new Date().toISOString()) {
|
|
69
|
-
const parsed =
|
|
69
|
+
const parsed = parseEvalTrialRef(trial, "eval rerun trial");
|
|
70
70
|
if (parsed.observation_status !== "valid")
|
|
71
71
|
throw new TypeError("eval rerun replacement must be valid");
|
|
72
72
|
const key = evalTrialKey(parsed);
|
|
@@ -118,7 +118,7 @@ export function parseEvalProgress(value) {
|
|
|
118
118
|
}
|
|
119
119
|
if (!Array.isArray(record.trials))
|
|
120
120
|
throw new TypeError("eval progress trials are invalid");
|
|
121
|
-
const trials = record.trials.map((trial, index) =>
|
|
121
|
+
const trials = record.trials.map((trial, index) => parseEvalTrialRef(trial, `eval progress trial ${index}`));
|
|
122
122
|
if (new Set(trials.map((trial) => trial.trial_id)).size !== trials.length
|
|
123
123
|
|| new Set(trials.map((trial) => trial.run_id)).size !== trials.length)
|
|
124
124
|
throw new TypeError("eval progress trial identities are duplicated");
|
|
@@ -164,7 +164,7 @@ export function parseEvalProgress(value) {
|
|
|
164
164
|
updated_at: record.updated_at,
|
|
165
165
|
};
|
|
166
166
|
}
|
|
167
|
-
function
|
|
167
|
+
export function parseEvalTrialRef(value, label = "eval trial") {
|
|
168
168
|
if (!value || typeof value !== "object" || Array.isArray(value))
|
|
169
169
|
throw new TypeError(`${label} must be an object`);
|
|
170
170
|
const trial = value;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"progress.js","sourceRoot":"","sources":["../../../src/evals/progress.ts"],"names":[],"mappings":"AAAA,OAAO,IAAI,MAAM,WAAW,CAAC;AAE7B,OAAO,EAAE,eAAe,EAAE,QAAQ,EAAE,MAAM,wBAAwB,CAAC;AAEnE,MAAM,OAAO,GAAG,qBAAqB,CAAC;AACtC,MAAM,MAAM,GAAG,oBAAoB,CAAC;AAEpC,MAAM,UAAU,kBAAkB,CAAC,KAOlC;IACC,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC;QAAE,MAAM,IAAI,SAAS,CAAC,kCAAkC,CAAC,CAAC;IACzF,IAAI,CAAC,KAAK,CAAC,WAAW,IAAI,CAAC,KAAK,CAAC,iBAAiB;QAAE,MAAM,IAAI,SAAS,CAAC,6CAA6C,CAAC,CAAC;IACvH,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;QAAE,MAAM,IAAI,SAAS,CAAC,qCAAqC,CAAC,CAAC;IAC9G,OAAO;QACL,cAAc,EAAE,GAAG;QACnB,OAAO,EAAE,KAAK,CAAC,MAAM;QACrB,YAAY,EAAE,KAAK,CAAC,WAAW;QAC/B,kBAAkB,EAAE,KAAK,CAAC,iBAAiB;QAC3C,MAAM,EAAE,SAAS;QACjB,UAAU,EAAE,CAAC;QACb,aAAa,EAAE,KAAK,CAAC,YAAY,IAAI,IAAI;QACzC,cAAc,EAAE,KAAK,CAAC,aAAa,IAAI,IAAI;QAC3C,MAAM,EAAE,EAAE;QACV,OAAO,EAAE,EAAE,cAAc,EAAE,CAAC,EAAE,YAAY,EAAE,CAAC,EAAE,cAAc,EAAE,CAAC,EAAE;QAClE,UAAU,EAAE,KAAK,CAAC,SAAS;QAC3B,UAAU,EAAE,KAAK,CAAC,SAAS;KAC5B,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,gBAAgB,CAAC,aAAqB;IAC1D,MAAM,KAAK,GAAG,MAAM,QAAQ,CAAiB,IAAI,CAAC,IAAI,CAAC,aAAa,EAAE,eAAe,CAAC,EAAE,IAAI,CAAC,CAAC;IAC9F,OAAO,KAAK,KAAK,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,iBAAiB,CAAC,KAAK,CAAC,CAAC;AAC1D,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,iBAAiB,CAAC,aAAqB,EAAE,QAAwB;IACrF,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,aAAa,EAAE,eAAe,CAAC,EAAE,iBAAiB,CAAC,QAAQ,CAAC,CAAC,CAAC;AAChG,CAAC;AAED,MAAM,UAAU,sBAAsB,CAAC,QAAwB,EAAE,KAAqB,EAAE,GAAG,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;IACpH,MAAM,MAAM,GAAG,UAAU,CAAC,KAAK,EAAE,qBAAqB,CAAC,CAAC;IACxD,MAAM,OAAO,GAAG,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,QAAQ,KAAK,MAAM,CAAC,QAAQ,CAAC,CAAC;IAClF,IAAI,OAAO,KAAK,SAAS,EAAE,CAAC;QAC1B,IAAI,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,KAAK,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC;YAAE,MAAM,IAAI,SAAS,CAAC,0CAA0C,MAAM,CAAC,QAAQ,EAAE,CAAC,CAAC;QACzI,OAAO,QAAQ,CAAC;IAClB,CAAC;IACD,IAAI,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,MAAM,KAAK,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC;QAClE,MAAM,IAAI,SAAS,CAAC,wCAAwC,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC;IAC/E,CAAC;IACD,IAAI,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,YAAY,CAAC,IAAI,CAAC,KAAK,YAAY,CAAC,MAAM,CAAC,CAAC,EAAE,CAAC;QAChF,MAAM,IAAI,SAAS,CAAC,yCAAyC,MAAM,CAAC,OAAO,YAAY,MAAM,CAAC,OAAO,EAAE,CAAC,CAAC;IAC3G,CAAC;IACD,MAAM,MAAM,GAAG,CAAC,GAAG,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,OAAO,CAAC,aAAa,CAAC,KAAK,CAAC,OAAO,CAAC;WACtG,IAAI,CAAC,OAAO,GAAG,KAAK,CAAC,OAAO;WAC5B,IAAI,CAAC,QAAQ,CAAC,aAAa,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC;IAClD,MAAM,KAAK,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,kBAAkB,KAAK,OAAO,CAAC,CAAC,MAAM,CAAC;IAClF,OAAO,iBAAiB,CAAC;QACvB,GAAG,QAAQ;QACX,UAAU,EAAE,QAAQ,CAAC,UAAU,GAAG,CAAC;QACnC,MAAM;QACN,OAAO,EAAE;YACP,cAAc,EAAE,MAAM,CAAC,MAAM;YAC7B,YAAY,EAAE,KAAK;YACnB,cAAc,EAAE,MAAM,CAAC,MAAM,GAAG,KAAK;SACtC;QACD,UAAU,EAAE,GAAG;KAChB,CAAC,CAAC;AACL,CAAC;AAED,MAAM,UAAU,YAAY,CAAC,KAAkD;IAC7E,OAAO,GAAG,KAAK,CAAC,OAAO,SAAS,KAAK,CAAC,OAAO,EAAE,CAAC;AAClD,CAAC;AAED,+FAA+F;AAC/F,MAAM,UAAU,+BAA+B,CAC7C,QAAwB,EACxB,KAAqB,EACrB,GAAG,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;IAE9B,MAAM,MAAM,GAAG,UAAU,CAAC,KAAK,EAAE,kBAAkB,CAAC,CAAC;IACrD,IAAI,MAAM,CAAC,kBAAkB,KAAK,OAAO;QAAE,MAAM,IAAI,SAAS,CAAC,sCAAsC,CAAC,CAAC;IACvG,MAAM,GAAG,GAAG,YAAY,CAAC,MAAM,CAAC,CAAC;IACjC,MAAM,QAAQ,GAAG,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,YAAY,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,CAAC;IAC5E,IAAI,QAAQ,EAAE,kBAAkB,KAAK,OAAO,EAAE,CAAC;QAC7C,IAAI,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,KAAK,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC;YAAE,OAAO,QAAQ,CAAC;QACzE,MAAM,IAAI,SAAS,CAAC,yCAAyC,MAAM,CAAC,OAAO,EAAE,CAAC,CAAC;IACjF,CAAC;IACD,IAAI,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,MAAM,KAAK,MAAM,CAAC,MAAM,IAAI,YAAY,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,EAAE,CAAC;QAChG,MAAM,IAAI,SAAS,CAAC,wCAAwC,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC;IAC/E,CAAC;IACD,IAAI,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,QAAQ,KAAK,MAAM,CAAC,QAAQ,IAAI,YAAY,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,EAAE,CAAC;QACpG,MAAM,IAAI,SAAS,CAAC,0CAA0C,MAAM,CAAC,QAAQ,EAAE,CAAC,CAAC;IACnF,CAAC;IACD,MAAM,MAAM,GAAG,CAAC,GAAG,QAAQ,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,YAAY,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,EAAE,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,OAAO,CAAC,aAAa,CAAC,KAAK,CAAC,OAAO,CAAC;WACnJ,IAAI,CAAC,OAAO,GAAG,KAAK,CAAC,OAAO;WAC5B,IAAI,CAAC,QAAQ,CAAC,aAAa,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC;IAClD,MAAM,KAAK,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,kBAAkB,KAAK,OAAO,CAAC,CAAC,MAAM,CAAC;IAClF,OAAO,iBAAiB,CAAC;QACvB,GAAG,QAAQ;QACX,UAAU,EAAE,QAAQ,CAAC,UAAU,GAAG,CAAC;QACnC,MAAM;QACN,OAAO,EAAE;YACP,cAAc,EAAE,MAAM,CAAC,MAAM;YAC7B,YAAY,EAAE,KAAK;YACnB,cAAc,EAAE,MAAM,CAAC,MAAM,GAAG,KAAK;SACtC;QACD,UAAU,EAAE,GAAG;KAChB,CAAC,CAAC;AACL,CAAC;AAED,MAAM,UAAU,iBAAiB,CAAC,KAAc;IAC9C,IAAI,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;QAAE,MAAM,IAAI,SAAS,CAAC,iCAAiC,CAAC,CAAC;IACxH,MAAM,MAAM,GAAG,KAAgC,CAAC;IAChD,IAAI,MAAM,CAAC,cAAc,KAAK,GAAG,IAAI,OAAO,MAAM,CAAC,OAAO,KAAK,QAAQ,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC;QACzG,MAAM,IAAI,SAAS,CAAC,mCAAmC,CAAC,CAAC;IAC3D,CAAC;IACD,IAAI,OAAO,MAAM,CAAC,YAAY,KAAK,QAAQ,IAAI,CAAC,MAAM,CAAC,YAAY;WAC9D,OAAO,MAAM,CAAC,kBAAkB,KAAK,QAAQ,IAAI,CAAC,MAAM,CAAC,kBAAkB;WAC3E,MAAM,CAAC,MAAM,KAAK,SAAS;QAAE,MAAM,IAAI,SAAS,CAAC,2CAA2C,CAAC,CAAC;IACnG,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,MAAM,CAAC,UAAU,CAAC,IAAK,MAAM,CAAC,UAAqB,GAAG,CAAC;QAAE,MAAM,IAAI,SAAS,CAAC,qCAAqC,CAAC,CAAC;IAC9I,KAAK,MAAM,IAAI,IAAI,CAAC,eAAe,EAAE,gBAAgB,CAAU,EAAE,CAAC;QAChE,MAAM,OAAO,GAAG,MAAM,CAAC,IAAI,CAAC,CAAC;QAC7B,IAAI,OAAO,KAAK,IAAI,IAAI,CAAC,CAAC,MAAM,CAAC,aAAa,CAAC,OAAO,CAAC,IAAK,OAAkB,GAAG,CAAC,CAAC;YAAE,MAAM,IAAI,SAAS,CAAC,iBAAiB,IAAI,aAAa,CAAC,CAAC;IAC/I,CAAC;IACD,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC;QAAE,MAAM,IAAI,SAAS,CAAC,kCAAkC,CAAC,CAAC;IAC3F,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,KAAK,EAAE,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,uBAAuB,KAAK,EAAE,CAAC,CAAC,CAAC;IACtG,IAAI,IAAI,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC,MAAM;WACpE,IAAI,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC,MAAM;QAAE,MAAM,IAAI,SAAS,CAAC,+CAA+C,CAAC,CAAC;IAC/I,MAAM,MAAM,GAAG,CAAC,GAAG,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,OAAO,CAAC,aAAa,CAAC,KAAK,CAAC,OAAO,CAAC;WACrF,IAAI,CAAC,OAAO,GAAG,KAAK,CAAC,OAAO;WAC5B,IAAI,CAAC,QAAQ,CAAC,aAAa,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC;IAClD,IAAI,MAAM,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,KAAK,EAAE,EAAE,CAAC,KAAK,KAAK,MAAM,CAAC,KAAK,CAAC,CAAC;QAAE,MAAM,IAAI,SAAS,CAAC,iDAAiD,CAAC,CAAC;IACnI,MAAM,OAAO,GAAG,MAAM,CAAC,OAAO,CAAC;IAC/B,IAAI,CAAC,OAAO,IAAI,OAAO,OAAO,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC;QAAE,MAAM,IAAI,SAAS,CAAC,kCAAkC,CAAC,CAAC;IAC/H,MAAM,aAAa,GAAG,OAAkC,CAAC;IACzD,MAAM,KAAK,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,kBAAkB,KAAK,OAAO,CAAC,CAAC,MAAM,CAAC;IACpF,IAAI,MAAM,CAAC,cAAc,KAAK,IAAI,IAAI,MAAM,CAAC,MAAM,GAAI,MAAM,CAAC,cAAyB,EAAE,CAAC;QACxF,MAAM,IAAI,SAAS,CAAC,oDAAoD,CAAC,CAAC;IAC5E,CAAC;IACD,IAAI,MAAM,CAAC,aAAa,KAAK,IAAI,IAAI,IAAI,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,GAAI,MAAM,CAAC,aAAwB,EAAE,CAAC;QAC3H,MAAM,IAAI,SAAS,CAAC,mDAAmD,CAAC,CAAC;IAC3E,CAAC;IACD,IAAI,aAAa,CAAC,cAAc,KAAK,MAAM,CAAC,MAAM,IAAI,aAAa,CAAC,YAAY,KAAK,KAAK;WACrF,aAAa,CAAC,cAAc,KAAK,MAAM,CAAC,MAAM,GAAG,KAAK;QAAE,MAAM,IAAI,SAAS,CAAC,6CAA6C,CAAC,CAAC;IAChI,IAAI,OAAO,MAAM,CAAC,UAAU,KAAK,QAAQ,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC;WACvF,OAAO,MAAM,CAAC,UAAU,KAAK,QAAQ,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,EAAE,CAAC;QAC9F,MAAM,IAAI,SAAS,CAAC,sCAAsC,CAAC,CAAC;IAC9D,CAAC;IACD,OAAO;QACL,cAAc,EAAE,GAAG;QACnB,OAAO,EAAE,MAAM,CAAC,OAAO;QACvB,YAAY,EAAE,MAAM,CAAC,YAAY;QACjC,kBAAkB,EAAE,MAAM,CAAC,kBAAkB;QAC7C,MAAM,EAAE,SAAS;QACjB,UAAU,EAAE,MAAM,CAAC,UAAoB;QACvC,aAAa,EAAE,MAAM,CAAC,aAA8B;QACpD,cAAc,EAAE,MAAM,CAAC,cAA+B;QACtD,MAAM;QACN,OAAO,EAAE;YACP,cAAc,EAAE,MAAM,CAAC,MAAM;YAC7B,YAAY,EAAE,KAAK;YACnB,cAAc,EAAE,MAAM,CAAC,MAAM,GAAG,KAAK;SACtC;QACD,UAAU,EAAE,MAAM,CAAC,UAAU;QAC7B,UAAU,EAAE,MAAM,CAAC,UAAU;KAC9B,CAAC;AACJ,CAAC;AAED,SAAS,UAAU,CAAC,KAAc,EAAE,KAAa;IAC/C,IAAI,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;QAAE,MAAM,IAAI,SAAS,CAAC,GAAG,KAAK,oBAAoB,CAAC,CAAC;IACnH,MAAM,KAAK,GAAG,KAAgC,CAAC;IAC/C,IAAI,OAAO,KAAK,CAAC,QAAQ,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,QAAQ;WACpD,OAAO,KAAK,CAAC,MAAM,KAAK,QAAQ,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC;WAC9D,OAAO,KAAK,CAAC,OAAO,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO;WACnD,CAAC,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,OAAO,CAAC,IAAK,KAAK,CAAC,OAAkB,GAAG,CAAC;WACrE,CAAC,KAAK,CAAC,kBAAkB,KAAK,OAAO,IAAI,KAAK,CAAC,kBAAkB,KAAK,SAAS,CAAC,EAAE,CAAC;QACtF,MAAM,IAAI,SAAS,CAAC,GAAG,KAAK,sBAAsB,CAAC,CAAC;IACtD,CAAC;IACD,IAAI,KAAK,CAAC,kBAAkB,KAAK,OAAO,IAAI,CAAC,OAAO,KAAK,CAAC,MAAM,KAAK,QAAQ,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,EAAE,CAAC;QACjH,MAAM,IAAI,SAAS,CAAC,GAAG,KAAK,0BAA0B,CAAC,CAAC;IAC1D,CAAC;IACD,IAAI,KAAK,CAAC,kBAAkB,KAAK,SAAS,IAAI,CAAC,OAAO,KAAK,CAAC,cAAc,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,cAAc,CAAC,EAAE,CAAC;QAClH,MAAM,IAAI,SAAS,CAAC,GAAG,KAAK,4BAA4B,CAAC,CAAC;IAC5D,CAAC;IACD,IAAI,KAAK,CAAC,mBAAmB,KAAK,SAAS,IAAI,CAAC,OAAO,KAAK,CAAC,mBAAmB,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,mBAAmB,CAAC,EAAE,CAAC;QAC7H,MAAM,IAAI,SAAS,CAAC,GAAG,KAAK,0BAA0B,CAAC,CAAC;IAC1D,CAAC;IACD,OAAO;QACL,QAAQ,EAAE,KAAK,CAAC,QAAQ;QACxB,MAAM,EAAE,KAAK,CAAC,MAAM;QACpB,OAAO,EAAE,KAAK,CAAC,OAAO;QACtB,OAAO,EAAE,KAAK,CAAC,OAAiB;QAChC,kBAAkB,EAAE,KAAK,CAAC,kBAAkB;QAC5C,GAAG,CAAC,KAAK,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,KAAK,CAAC,MAAgB,EAAE,CAAC;QACzE,GAAG,CAAC,KAAK,CAAC,mBAAmB,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,mBAAmB,EAAE,KAAK,CAAC,mBAA6B,EAAE,CAAC;QAChH,GAAG,CAAC,KAAK,CAAC,cAAc,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,cAAc,EAAE,KAAK,CAAC,cAAwB,EAAE,CAAC;KAClG,CAAC;AACJ,CAAC"}
|
|
1
|
+
{"version":3,"file":"progress.js","sourceRoot":"","sources":["../../../src/evals/progress.ts"],"names":[],"mappings":"AAAA,OAAO,IAAI,MAAM,WAAW,CAAC;AAE7B,OAAO,EAAE,eAAe,EAAE,QAAQ,EAAE,MAAM,wBAAwB,CAAC;AAEnE,MAAM,OAAO,GAAG,qBAAqB,CAAC;AACtC,MAAM,MAAM,GAAG,oBAAoB,CAAC;AAEpC,MAAM,UAAU,kBAAkB,CAAC,KAOlC;IACC,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC;QAAE,MAAM,IAAI,SAAS,CAAC,kCAAkC,CAAC,CAAC;IACzF,IAAI,CAAC,KAAK,CAAC,WAAW,IAAI,CAAC,KAAK,CAAC,iBAAiB;QAAE,MAAM,IAAI,SAAS,CAAC,6CAA6C,CAAC,CAAC;IACvH,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;QAAE,MAAM,IAAI,SAAS,CAAC,qCAAqC,CAAC,CAAC;IAC9G,OAAO;QACL,cAAc,EAAE,GAAG;QACnB,OAAO,EAAE,KAAK,CAAC,MAAM;QACrB,YAAY,EAAE,KAAK,CAAC,WAAW;QAC/B,kBAAkB,EAAE,KAAK,CAAC,iBAAiB;QAC3C,MAAM,EAAE,SAAS;QACjB,UAAU,EAAE,CAAC;QACb,aAAa,EAAE,KAAK,CAAC,YAAY,IAAI,IAAI;QACzC,cAAc,EAAE,KAAK,CAAC,aAAa,IAAI,IAAI;QAC3C,MAAM,EAAE,EAAE;QACV,OAAO,EAAE,EAAE,cAAc,EAAE,CAAC,EAAE,YAAY,EAAE,CAAC,EAAE,cAAc,EAAE,CAAC,EAAE;QAClE,UAAU,EAAE,KAAK,CAAC,SAAS;QAC3B,UAAU,EAAE,KAAK,CAAC,SAAS;KAC5B,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,gBAAgB,CAAC,aAAqB;IAC1D,MAAM,KAAK,GAAG,MAAM,QAAQ,CAAiB,IAAI,CAAC,IAAI,CAAC,aAAa,EAAE,eAAe,CAAC,EAAE,IAAI,CAAC,CAAC;IAC9F,OAAO,KAAK,KAAK,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,iBAAiB,CAAC,KAAK,CAAC,CAAC;AAC1D,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,iBAAiB,CAAC,aAAqB,EAAE,QAAwB;IACrF,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,aAAa,EAAE,eAAe,CAAC,EAAE,iBAAiB,CAAC,QAAQ,CAAC,CAAC,CAAC;AAChG,CAAC;AAED,MAAM,UAAU,sBAAsB,CAAC,QAAwB,EAAE,KAAqB,EAAE,GAAG,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;IACpH,MAAM,MAAM,GAAG,iBAAiB,CAAC,KAAK,EAAE,qBAAqB,CAAC,CAAC;IAC/D,MAAM,OAAO,GAAG,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,QAAQ,KAAK,MAAM,CAAC,QAAQ,CAAC,CAAC;IAClF,IAAI,OAAO,KAAK,SAAS,EAAE,CAAC;QAC1B,IAAI,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,KAAK,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC;YAAE,MAAM,IAAI,SAAS,CAAC,0CAA0C,MAAM,CAAC,QAAQ,EAAE,CAAC,CAAC;QACzI,OAAO,QAAQ,CAAC;IAClB,CAAC;IACD,IAAI,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,MAAM,KAAK,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC;QAClE,MAAM,IAAI,SAAS,CAAC,wCAAwC,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC;IAC/E,CAAC;IACD,IAAI,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,YAAY,CAAC,IAAI,CAAC,KAAK,YAAY,CAAC,MAAM,CAAC,CAAC,EAAE,CAAC;QAChF,MAAM,IAAI,SAAS,CAAC,yCAAyC,MAAM,CAAC,OAAO,YAAY,MAAM,CAAC,OAAO,EAAE,CAAC,CAAC;IAC3G,CAAC;IACD,MAAM,MAAM,GAAG,CAAC,GAAG,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,OAAO,CAAC,aAAa,CAAC,KAAK,CAAC,OAAO,CAAC;WACtG,IAAI,CAAC,OAAO,GAAG,KAAK,CAAC,OAAO;WAC5B,IAAI,CAAC,QAAQ,CAAC,aAAa,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC;IAClD,MAAM,KAAK,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,kBAAkB,KAAK,OAAO,CAAC,CAAC,MAAM,CAAC;IAClF,OAAO,iBAAiB,CAAC;QACvB,GAAG,QAAQ;QACX,UAAU,EAAE,QAAQ,CAAC,UAAU,GAAG,CAAC;QACnC,MAAM;QACN,OAAO,EAAE;YACP,cAAc,EAAE,MAAM,CAAC,MAAM;YAC7B,YAAY,EAAE,KAAK;YACnB,cAAc,EAAE,MAAM,CAAC,MAAM,GAAG,KAAK;SACtC;QACD,UAAU,EAAE,GAAG;KAChB,CAAC,CAAC;AACL,CAAC;AAED,MAAM,UAAU,YAAY,CAAC,KAAkD;IAC7E,OAAO,GAAG,KAAK,CAAC,OAAO,SAAS,KAAK,CAAC,OAAO,EAAE,CAAC;AAClD,CAAC;AAED,+FAA+F;AAC/F,MAAM,UAAU,+BAA+B,CAC7C,QAAwB,EACxB,KAAqB,EACrB,GAAG,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;IAE9B,MAAM,MAAM,GAAG,iBAAiB,CAAC,KAAK,EAAE,kBAAkB,CAAC,CAAC;IAC5D,IAAI,MAAM,CAAC,kBAAkB,KAAK,OAAO;QAAE,MAAM,IAAI,SAAS,CAAC,sCAAsC,CAAC,CAAC;IACvG,MAAM,GAAG,GAAG,YAAY,CAAC,MAAM,CAAC,CAAC;IACjC,MAAM,QAAQ,GAAG,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,YAAY,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,CAAC;IAC5E,IAAI,QAAQ,EAAE,kBAAkB,KAAK,OAAO,EAAE,CAAC;QAC7C,IAAI,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,KAAK,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC;YAAE,OAAO,QAAQ,CAAC;QACzE,MAAM,IAAI,SAAS,CAAC,yCAAyC,MAAM,CAAC,OAAO,EAAE,CAAC,CAAC;IACjF,CAAC;IACD,IAAI,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,MAAM,KAAK,MAAM,CAAC,MAAM,IAAI,YAAY,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,EAAE,CAAC;QAChG,MAAM,IAAI,SAAS,CAAC,wCAAwC,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC;IAC/E,CAAC;IACD,IAAI,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,QAAQ,KAAK,MAAM,CAAC,QAAQ,IAAI,YAAY,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,EAAE,CAAC;QACpG,MAAM,IAAI,SAAS,CAAC,0CAA0C,MAAM,CAAC,QAAQ,EAAE,CAAC,CAAC;IACnF,CAAC;IACD,MAAM,MAAM,GAAG,CAAC,GAAG,QAAQ,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,YAAY,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,EAAE,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,OAAO,CAAC,aAAa,CAAC,KAAK,CAAC,OAAO,CAAC;WACnJ,IAAI,CAAC,OAAO,GAAG,KAAK,CAAC,OAAO;WAC5B,IAAI,CAAC,QAAQ,CAAC,aAAa,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC;IAClD,MAAM,KAAK,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,kBAAkB,KAAK,OAAO,CAAC,CAAC,MAAM,CAAC;IAClF,OAAO,iBAAiB,CAAC;QACvB,GAAG,QAAQ;QACX,UAAU,EAAE,QAAQ,CAAC,UAAU,GAAG,CAAC;QACnC,MAAM;QACN,OAAO,EAAE;YACP,cAAc,EAAE,MAAM,CAAC,MAAM;YAC7B,YAAY,EAAE,KAAK;YACnB,cAAc,EAAE,MAAM,CAAC,MAAM,GAAG,KAAK;SACtC;QACD,UAAU,EAAE,GAAG;KAChB,CAAC,CAAC;AACL,CAAC;AAED,MAAM,UAAU,iBAAiB,CAAC,KAAc;IAC9C,IAAI,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;QAAE,MAAM,IAAI,SAAS,CAAC,iCAAiC,CAAC,CAAC;IACxH,MAAM,MAAM,GAAG,KAAgC,CAAC;IAChD,IAAI,MAAM,CAAC,cAAc,KAAK,GAAG,IAAI,OAAO,MAAM,CAAC,OAAO,KAAK,QAAQ,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC;QACzG,MAAM,IAAI,SAAS,CAAC,mCAAmC,CAAC,CAAC;IAC3D,CAAC;IACD,IAAI,OAAO,MAAM,CAAC,YAAY,KAAK,QAAQ,IAAI,CAAC,MAAM,CAAC,YAAY;WAC9D,OAAO,MAAM,CAAC,kBAAkB,KAAK,QAAQ,IAAI,CAAC,MAAM,CAAC,kBAAkB;WAC3E,MAAM,CAAC,MAAM,KAAK,SAAS;QAAE,MAAM,IAAI,SAAS,CAAC,2CAA2C,CAAC,CAAC;IACnG,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,MAAM,CAAC,UAAU,CAAC,IAAK,MAAM,CAAC,UAAqB,GAAG,CAAC;QAAE,MAAM,IAAI,SAAS,CAAC,qCAAqC,CAAC,CAAC;IAC9I,KAAK,MAAM,IAAI,IAAI,CAAC,eAAe,EAAE,gBAAgB,CAAU,EAAE,CAAC;QAChE,MAAM,OAAO,GAAG,MAAM,CAAC,IAAI,CAAC,CAAC;QAC7B,IAAI,OAAO,KAAK,IAAI,IAAI,CAAC,CAAC,MAAM,CAAC,aAAa,CAAC,OAAO,CAAC,IAAK,OAAkB,GAAG,CAAC,CAAC;YAAE,MAAM,IAAI,SAAS,CAAC,iBAAiB,IAAI,aAAa,CAAC,CAAC;IAC/I,CAAC;IACD,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC;QAAE,MAAM,IAAI,SAAS,CAAC,kCAAkC,CAAC,CAAC;IAC3F,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,KAAK,EAAE,EAAE,CAAC,iBAAiB,CAAC,KAAK,EAAE,uBAAuB,KAAK,EAAE,CAAC,CAAC,CAAC;IAC7G,IAAI,IAAI,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC,MAAM;WACpE,IAAI,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC,MAAM;QAAE,MAAM,IAAI,SAAS,CAAC,+CAA+C,CAAC,CAAC;IAC/I,MAAM,MAAM,GAAG,CAAC,GAAG,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,OAAO,CAAC,aAAa,CAAC,KAAK,CAAC,OAAO,CAAC;WACrF,IAAI,CAAC,OAAO,GAAG,KAAK,CAAC,OAAO;WAC5B,IAAI,CAAC,QAAQ,CAAC,aAAa,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC;IAClD,IAAI,MAAM,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,KAAK,EAAE,EAAE,CAAC,KAAK,KAAK,MAAM,CAAC,KAAK,CAAC,CAAC;QAAE,MAAM,IAAI,SAAS,CAAC,iDAAiD,CAAC,CAAC;IACnI,MAAM,OAAO,GAAG,MAAM,CAAC,OAAO,CAAC;IAC/B,IAAI,CAAC,OAAO,IAAI,OAAO,OAAO,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC;QAAE,MAAM,IAAI,SAAS,CAAC,kCAAkC,CAAC,CAAC;IAC/H,MAAM,aAAa,GAAG,OAAkC,CAAC;IACzD,MAAM,KAAK,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,kBAAkB,KAAK,OAAO,CAAC,CAAC,MAAM,CAAC;IACpF,IAAI,MAAM,CAAC,cAAc,KAAK,IAAI,IAAI,MAAM,CAAC,MAAM,GAAI,MAAM,CAAC,cAAyB,EAAE,CAAC;QACxF,MAAM,IAAI,SAAS,CAAC,oDAAoD,CAAC,CAAC;IAC5E,CAAC;IACD,IAAI,MAAM,CAAC,aAAa,KAAK,IAAI,IAAI,IAAI,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,GAAI,MAAM,CAAC,aAAwB,EAAE,CAAC;QAC3H,MAAM,IAAI,SAAS,CAAC,mDAAmD,CAAC,CAAC;IAC3E,CAAC;IACD,IAAI,aAAa,CAAC,cAAc,KAAK,MAAM,CAAC,MAAM,IAAI,aAAa,CAAC,YAAY,KAAK,KAAK;WACrF,aAAa,CAAC,cAAc,KAAK,MAAM,CAAC,MAAM,GAAG,KAAK;QAAE,MAAM,IAAI,SAAS,CAAC,6CAA6C,CAAC,CAAC;IAChI,IAAI,OAAO,MAAM,CAAC,UAAU,KAAK,QAAQ,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC;WACvF,OAAO,MAAM,CAAC,UAAU,KAAK,QAAQ,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,EAAE,CAAC;QAC9F,MAAM,IAAI,SAAS,CAAC,sCAAsC,CAAC,CAAC;IAC9D,CAAC;IACD,OAAO;QACL,cAAc,EAAE,GAAG;QACnB,OAAO,EAAE,MAAM,CAAC,OAAO;QACvB,YAAY,EAAE,MAAM,CAAC,YAAY;QACjC,kBAAkB,EAAE,MAAM,CAAC,kBAAkB;QAC7C,MAAM,EAAE,SAAS;QACjB,UAAU,EAAE,MAAM,CAAC,UAAoB;QACvC,aAAa,EAAE,MAAM,CAAC,aAA8B;QACpD,cAAc,EAAE,MAAM,CAAC,cAA+B;QACtD,MAAM;QACN,OAAO,EAAE;YACP,cAAc,EAAE,MAAM,CAAC,MAAM;YAC7B,YAAY,EAAE,KAAK;YACnB,cAAc,EAAE,MAAM,CAAC,MAAM,GAAG,KAAK;SACtC;QACD,UAAU,EAAE,MAAM,CAAC,UAAU;QAC7B,UAAU,EAAE,MAAM,CAAC,UAAU;KAC9B,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,iBAAiB,CAAC,KAAc,EAAE,KAAK,GAAG,YAAY;IACpE,IAAI,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;QAAE,MAAM,IAAI,SAAS,CAAC,GAAG,KAAK,oBAAoB,CAAC,CAAC;IACnH,MAAM,KAAK,GAAG,KAAgC,CAAC;IAC/C,IAAI,OAAO,KAAK,CAAC,QAAQ,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,QAAQ;WACpD,OAAO,KAAK,CAAC,MAAM,KAAK,QAAQ,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC;WAC9D,OAAO,KAAK,CAAC,OAAO,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO;WACnD,CAAC,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,OAAO,CAAC,IAAK,KAAK,CAAC,OAAkB,GAAG,CAAC;WACrE,CAAC,KAAK,CAAC,kBAAkB,KAAK,OAAO,IAAI,KAAK,CAAC,kBAAkB,KAAK,SAAS,CAAC,EAAE,CAAC;QACtF,MAAM,IAAI,SAAS,CAAC,GAAG,KAAK,sBAAsB,CAAC,CAAC;IACtD,CAAC;IACD,IAAI,KAAK,CAAC,kBAAkB,KAAK,OAAO,IAAI,CAAC,OAAO,KAAK,CAAC,MAAM,KAAK,QAAQ,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,EAAE,CAAC;QACjH,MAAM,IAAI,SAAS,CAAC,GAAG,KAAK,0BAA0B,CAAC,CAAC;IAC1D,CAAC;IACD,IAAI,KAAK,CAAC,kBAAkB,KAAK,SAAS,IAAI,CAAC,OAAO,KAAK,CAAC,cAAc,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,cAAc,CAAC,EAAE,CAAC;QAClH,MAAM,IAAI,SAAS,CAAC,GAAG,KAAK,4BAA4B,CAAC,CAAC;IAC5D,CAAC;IACD,IAAI,KAAK,CAAC,mBAAmB,KAAK,SAAS,IAAI,CAAC,OAAO,KAAK,CAAC,mBAAmB,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,mBAAmB,CAAC,EAAE,CAAC;QAC7H,MAAM,IAAI,SAAS,CAAC,GAAG,KAAK,0BAA0B,CAAC,CAAC;IAC1D,CAAC;IACD,OAAO;QACL,QAAQ,EAAE,KAAK,CAAC,QAAQ;QACxB,MAAM,EAAE,KAAK,CAAC,MAAM;QACpB,OAAO,EAAE,KAAK,CAAC,OAAO;QACtB,OAAO,EAAE,KAAK,CAAC,OAAiB;QAChC,kBAAkB,EAAE,KAAK,CAAC,kBAAkB;QAC5C,GAAG,CAAC,KAAK,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,KAAK,CAAC,MAAgB,EAAE,CAAC;QACzE,GAAG,CAAC,KAAK,CAAC,mBAAmB,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,mBAAmB,EAAE,KAAK,CAAC,mBAA6B,EAAE,CAAC;QAChH,GAAG,CAAC,KAAK,CAAC,cAAc,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,cAAc,EAAE,KAAK,CAAC,cAAwB,EAAE,CAAC;KAClG,CAAC;AACJ,CAAC"}
|
|
@@ -1,7 +1,8 @@
|
|
|
1
|
-
import { readdir } from "node:fs/promises";
|
|
1
|
+
import { lstat, readdir } from "node:fs/promises";
|
|
2
2
|
import path from "node:path";
|
|
3
3
|
import { inspectEvalRuntimeKind } from "../controller-runtime/index.js";
|
|
4
4
|
import { HitchError, SCHEMA_VERSION, invalidInput, readJSON, statePaths } from "../foundation/index.js";
|
|
5
|
+
import { readExecutionLeases } from "./execution-leases.js";
|
|
5
6
|
export async function listEvals({ root }) {
|
|
6
7
|
const directory = statePaths(root).evals;
|
|
7
8
|
let entries;
|
|
@@ -19,16 +20,17 @@ export async function listEvals({ root }) {
|
|
|
19
20
|
continue;
|
|
20
21
|
const result = await readJSON(path.join(directory, entry.name, "result.json"), null).catch(() => null);
|
|
21
22
|
const request = await readJSON(path.join(directory, entry.name, "request.json"), null).catch(() => null);
|
|
23
|
+
const control = await readJSON(path.join(directory, entry.name, "control.json"), null).catch(() => null);
|
|
22
24
|
if (!result && !request)
|
|
23
25
|
continue;
|
|
24
26
|
evals.push({
|
|
25
27
|
eval_id: entry.name,
|
|
26
|
-
status: result?.status || "running",
|
|
28
|
+
status: result?.status || control?.state || "running",
|
|
27
29
|
backend: result?.backend?.name || request?.backend || null,
|
|
28
30
|
dataset: result?.dataset || request?.dataset || null,
|
|
29
31
|
harness_ref: result?.candidate?.harness_ref || request?.harness_ref || null,
|
|
30
32
|
primary_reward: result?.summary?.primary_reward ?? null,
|
|
31
|
-
started_at: result?.started_at || null,
|
|
33
|
+
started_at: result?.started_at || control?.created_at || null,
|
|
32
34
|
completed_at: result?.completed_at || null,
|
|
33
35
|
});
|
|
34
36
|
}
|
|
@@ -48,8 +50,23 @@ export async function inspectEval(evalId, { root }) {
|
|
|
48
50
|
request,
|
|
49
51
|
resolution: await readJSON(path.join(directory, "resolution.json"), null),
|
|
50
52
|
plan: await readJSON(path.join(directory, "plan.json"), null),
|
|
53
|
+
execution_plan: await readJSON(path.join(directory, "execution-plan.json"), null),
|
|
54
|
+
submission: await readJSON(path.join(directory, "submission.json"), null),
|
|
55
|
+
control: await readJSON(path.join(directory, "control.json"), null),
|
|
56
|
+
progress: await readJSON(path.join(directory, "progress.json"), null),
|
|
57
|
+
leases: await readExecutionLeases(directory),
|
|
51
58
|
result: await readJSON(path.join(directory, "result.json"), null),
|
|
52
59
|
runtime_storage: await inspectEvalRuntimeKind(directory),
|
|
53
60
|
};
|
|
54
61
|
}
|
|
62
|
+
export async function isControlPlaneEval(root, evalId) {
|
|
63
|
+
try {
|
|
64
|
+
return (await lstat(path.join(statePaths(root).evals, evalId, "submission.json"))).isFile();
|
|
65
|
+
}
|
|
66
|
+
catch (error) {
|
|
67
|
+
if (error.code === "ENOENT")
|
|
68
|
+
return false;
|
|
69
|
+
throw error;
|
|
70
|
+
}
|
|
71
|
+
}
|
|
55
72
|
//# sourceMappingURL=records.js.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"records.js","sourceRoot":"","sources":["../../../src/evals/records.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,kBAAkB,CAAC;
|
|
1
|
+
{"version":3,"file":"records.js","sourceRoot":"","sources":["../../../src/evals/records.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,EAAE,OAAO,EAAE,MAAM,kBAAkB,CAAC;AAClD,OAAO,IAAI,MAAM,WAAW,CAAC;AAC7B,OAAO,EAAE,sBAAsB,EAAE,MAAM,gCAAgC,CAAC;AAExE,OAAO,EAAE,UAAU,EAAE,cAAc,EAAE,YAAY,EAAE,QAAQ,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AACxG,OAAO,EAAE,mBAAmB,EAAE,MAAM,uBAAuB,CAAC;AAa5D,MAAM,CAAC,KAAK,UAAU,SAAS,CAAC,EAAE,IAAI,EAAoB;IACxD,MAAM,SAAS,GAAG,UAAU,CAAC,IAAI,CAAC,CAAC,KAAK,CAAC;IACzC,IAAI,OAAO,CAAC;IACZ,IAAI,CAAC;QACH,OAAO,GAAG,MAAM,OAAO,CAAC,SAAS,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC;IAC9D,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAK,KAA+B,EAAE,IAAI,KAAK,QAAQ;YAAE,OAAO,EAAE,CAAC;QACnE,MAAM,KAAK,CAAC;IACd,CAAC;IACD,MAAM,KAAK,GAAiB,EAAE,CAAC;IAC/B,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;QAC5B,IAAI,CAAC,KAAK,CAAC,WAAW,EAAE,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC,OAAO,CAAC;YAAE,SAAS;QACtE,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAiC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,KAAK,CAAC,IAAI,EAAE,aAAa,CAAC,EAAE,IAAI,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,CAAC;QACvI,MAAM,OAAO,GAAG,MAAM,QAAQ,CAAiC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,KAAK,CAAC,IAAI,EAAE,cAAc,CAAC,EAAE,IAAI,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,CAAC;QACzI,MAAM,OAAO,GAAG,MAAM,QAAQ,CAAiC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,KAAK,CAAC,IAAI,EAAE,cAAc,CAAC,EAAE,IAAI,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,CAAC;QACzI,IAAI,CAAC,MAAM,IAAI,CAAC,OAAO;YAAE,SAAS;QAClC,KAAK,CAAC,IAAI,CAAC;YACT,OAAO,EAAE,KAAK,CAAC,IAAI;YACnB,MAAM,EAAG,MAAM,EAAE,MAAiB,IAAK,OAAO,EAAE,KAAgB,IAAI,SAAS;YAC7E,OAAO,EAAI,MAAM,EAAE,OAAmC,EAAE,IAAe,IAAK,OAAO,EAAE,OAAkB,IAAI,IAAI;YAC/G,OAAO,EAAG,MAAM,EAAE,OAAkB,IAAK,OAAO,EAAE,OAAkB,IAAI,IAAI;YAC5E,WAAW,EAAI,MAAM,EAAE,SAAqC,EAAE,WAAsB,IAAK,OAAO,EAAE,WAAsB,IAAI,IAAI;YAChI,cAAc,EAAG,MAAM,EAAE,OAAmC,EAAE,cAA+B,IAAI,IAAI;YACrG,UAAU,EAAG,MAAM,EAAE,UAAqB,IAAK,OAAO,EAAE,UAAqB,IAAI,IAAI;YACrF,YAAY,EAAG,MAAM,EAAE,YAAuB,IAAI,IAAI;SACvD,CAAC,CAAC;IACL,CAAC;IACD,OAAO,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,MAAM,CAAC,KAAK,CAAC,UAAU,IAAI,KAAK,CAAC,OAAO,CAAC,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,UAAU,IAAI,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;AACvI,CAAC;AAkBD,MAAM,CAAC,KAAK,UAAU,WAAW,CAAC,MAAc,EAAE,EAAE,IAAI,EAAoB;IAC1E,IAAI,OAAO,MAAM,KAAK,QAAQ,IAAI,CAAC,qBAAqB,CAAC,IAAI,CAAC,MAAM,CAAC;QAAE,MAAM,YAAY,CAAC,oBAAoB,MAAM,EAAE,CAAC,CAAC;IACxH,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;IAC5D,MAAM,OAAO,GAAG,MAAM,QAAQ,CAAiC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,cAAc,CAAC,EAAE,IAAI,CAAC,CAAC;IAC3G,IAAI,CAAC,OAAO;QAAE,MAAM,IAAI,UAAU,CAAC,mBAAmB,MAAM,EAAE,EAAE,EAAE,IAAI,EAAE,gBAAgB,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC,CAAC;IACzG,OAAO;QACL,cAAc,EAAE,cAAc;QAC9B,OAAO,EAAE,MAAM;QACf,SAAS;QACT,OAAO;QACP,UAAU,EAAE,MAAM,QAAQ,CAAiC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,iBAAiB,CAAC,EAAE,IAAI,CAAC;QACzG,IAAI,EAAE,MAAM,QAAQ,CAAiC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,WAAW,CAAC,EAAE,IAAI,CAAC;QAC7F,cAAc,EAAE,MAAM,QAAQ,CAAiC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,qBAAqB,CAAC,EAAE,IAAI,CAAC;QACjH,UAAU,EAAE,MAAM,QAAQ,CAAiC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,iBAAiB,CAAC,EAAE,IAAI,CAAC;QACzG,OAAO,EAAE,MAAM,QAAQ,CAAiC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,cAAc,CAAC,EAAE,IAAI,CAAC;QACnG,QAAQ,EAAE,MAAM,QAAQ,CAAiC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,eAAe,CAAC,EAAE,IAAI,CAAC;QACrG,MAAM,EAAE,MAAM,mBAAmB,CAAC,SAAS,CAAC;QAC5C,MAAM,EAAE,MAAM,QAAQ,CAAiC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,aAAa,CAAC,EAAE,IAAI,CAAC;QACjG,eAAe,EAAE,MAAM,sBAAsB,CAAC,SAAS,CAAC;KACzD,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,kBAAkB,CAAC,IAAY,EAAE,MAAc;IACnE,IAAI,CAAC;QACH,OAAO,CAAC,MAAM,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,MAAM,EAAE,iBAAiB,CAAC,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC;IAC9F,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"}
|