agent-hitch 0.2.7 → 0.2.8
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 +5 -0
- 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 +41 -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 +19 -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 +27 -20
- 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 +541 -43
- 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 +1,219 @@
|
|
|
1
|
+
import { createHash } from "node:crypto";
|
|
2
|
+
import { lstat, readFile, readdir } from "node:fs/promises";
|
|
3
|
+
import path from "node:path";
|
|
4
|
+
import { invalidInput } from "../foundation/index.js";
|
|
5
|
+
const SHA256 = /^sha256:[0-9a-f]{64}$/;
|
|
6
|
+
/** Build the canonical manifest for an already-materialized Harbor dataset. */
|
|
7
|
+
export async function buildBenchmarkAdapterManifest(input) {
|
|
8
|
+
const ids = [...input.taskIds].sort();
|
|
9
|
+
if (!ids.length || new Set(ids).size !== ids.length)
|
|
10
|
+
throw invalidInput("benchmark adapter task IDs must be non-empty and unique");
|
|
11
|
+
const tasks = await Promise.all(ids.map(async (id) => {
|
|
12
|
+
const valid = taskId(id);
|
|
13
|
+
const directory = path.join(path.resolve(input.dataset), valid);
|
|
14
|
+
const info = await lstat(directory).catch((error) => {
|
|
15
|
+
if (error.code === "ENOENT")
|
|
16
|
+
throw invalidInput(`manifest task is missing: ${valid}`);
|
|
17
|
+
throw error;
|
|
18
|
+
});
|
|
19
|
+
if (!info.isDirectory() || info.isSymbolicLink())
|
|
20
|
+
throw invalidInput(`manifest task is not a directory: ${valid}`);
|
|
21
|
+
return { task_id: valid, task_digest: await taskTreeDigest(directory) };
|
|
22
|
+
}));
|
|
23
|
+
const body = {
|
|
24
|
+
schema_version: "1",
|
|
25
|
+
kind: "gear-harbor-benchmark",
|
|
26
|
+
benchmark: input.benchmark,
|
|
27
|
+
adapter: input.adapter,
|
|
28
|
+
scoring: input.scoring,
|
|
29
|
+
tasks,
|
|
30
|
+
};
|
|
31
|
+
return { ...body, dataset_digest: sha256(canonicalJson(body)) };
|
|
32
|
+
}
|
|
33
|
+
/** Load and integrity-check the optional Gear standardized-dataset manifest. */
|
|
34
|
+
export async function loadBenchmarkAdapterManifest(dataset) {
|
|
35
|
+
const root = path.resolve(dataset);
|
|
36
|
+
let raw;
|
|
37
|
+
try {
|
|
38
|
+
const info = await lstat(path.join(root, "benchmark.adapter.json"));
|
|
39
|
+
if (!info.isFile() || info.isSymbolicLink())
|
|
40
|
+
throw invalidInput("benchmark.adapter.json must be a regular file");
|
|
41
|
+
raw = JSON.parse(await readFile(path.join(root, "benchmark.adapter.json"), "utf8"));
|
|
42
|
+
}
|
|
43
|
+
catch (error) {
|
|
44
|
+
if (error.code === "ENOENT")
|
|
45
|
+
return null;
|
|
46
|
+
if (error instanceof SyntaxError)
|
|
47
|
+
throw invalidInput("benchmark.adapter.json is invalid JSON");
|
|
48
|
+
throw error;
|
|
49
|
+
}
|
|
50
|
+
const manifest = parseManifest(raw);
|
|
51
|
+
const actualTasks = [];
|
|
52
|
+
for (const task of manifest.tasks) {
|
|
53
|
+
const directory = path.join(root, task.task_id);
|
|
54
|
+
const info = await lstat(directory).catch((error) => {
|
|
55
|
+
if (error.code === "ENOENT")
|
|
56
|
+
throw invalidInput(`manifest task is missing: ${task.task_id}`);
|
|
57
|
+
throw error;
|
|
58
|
+
});
|
|
59
|
+
if (!info.isDirectory() || info.isSymbolicLink())
|
|
60
|
+
throw invalidInput(`manifest task is not a directory: ${task.task_id}`);
|
|
61
|
+
const digest = await taskTreeDigest(directory);
|
|
62
|
+
if (digest !== task.task_digest)
|
|
63
|
+
throw invalidInput(`manifest task digest mismatch: ${task.task_id}`);
|
|
64
|
+
actualTasks.push({ task_id: task.task_id, task_digest: digest });
|
|
65
|
+
}
|
|
66
|
+
const taskDirectories = [];
|
|
67
|
+
for (const entry of await readdir(root, { withFileTypes: true })) {
|
|
68
|
+
if (!entry.isDirectory())
|
|
69
|
+
continue;
|
|
70
|
+
try {
|
|
71
|
+
const taskInfo = await lstat(path.join(root, entry.name, "task.toml"));
|
|
72
|
+
if (taskInfo.isFile() && !taskInfo.isSymbolicLink())
|
|
73
|
+
taskDirectories.push(entry.name);
|
|
74
|
+
}
|
|
75
|
+
catch (error) {
|
|
76
|
+
if (error.code !== "ENOENT")
|
|
77
|
+
throw error;
|
|
78
|
+
}
|
|
79
|
+
}
|
|
80
|
+
if (JSON.stringify(taskDirectories.sort()) !== JSON.stringify(actualTasks.map((task) => task.task_id))) {
|
|
81
|
+
throw invalidInput("benchmark adapter manifest task set does not match the dataset");
|
|
82
|
+
}
|
|
83
|
+
const body = {
|
|
84
|
+
schema_version: manifest.schema_version,
|
|
85
|
+
kind: manifest.kind,
|
|
86
|
+
benchmark: manifest.benchmark,
|
|
87
|
+
adapter: manifest.adapter,
|
|
88
|
+
scoring: manifest.scoring,
|
|
89
|
+
tasks: actualTasks,
|
|
90
|
+
};
|
|
91
|
+
const expected = sha256(canonicalJson(body));
|
|
92
|
+
if (manifest.dataset_digest !== expected)
|
|
93
|
+
throw invalidInput("benchmark adapter dataset digest mismatch");
|
|
94
|
+
return manifest;
|
|
95
|
+
}
|
|
96
|
+
export function scoreWithinRange(score, definition) {
|
|
97
|
+
return score >= definition.range[0] && score <= definition.range[1];
|
|
98
|
+
}
|
|
99
|
+
function parseManifest(value) {
|
|
100
|
+
const record = object(value, "benchmark adapter manifest");
|
|
101
|
+
exact(record, ["schema_version", "kind", "benchmark", "adapter", "scoring", "tasks", "dataset_digest"], "benchmark adapter manifest");
|
|
102
|
+
if (record.schema_version !== "1" || record.kind !== "gear-harbor-benchmark")
|
|
103
|
+
throw invalidInput("unsupported benchmark adapter manifest");
|
|
104
|
+
const benchmark = namedRevision(record.benchmark, "benchmark");
|
|
105
|
+
const adapterRecord = object(record.adapter, "adapter");
|
|
106
|
+
exact(adapterRecord, ["id", "revision", "output_protocol"], "adapter");
|
|
107
|
+
if (adapterRecord.output_protocol !== "gear-harbor-eval-result-v1")
|
|
108
|
+
throw invalidInput("unsupported benchmark adapter output protocol");
|
|
109
|
+
const adapter = { id: identifier(adapterRecord.id, "adapter id"), revision: immutableRevision(adapterRecord.revision, "adapter revision"), output_protocol: adapterRecord.output_protocol };
|
|
110
|
+
const scoringRecord = object(record.scoring, "scoring");
|
|
111
|
+
exact(scoringRecord, ["total_score", "process_score"], "scoring");
|
|
112
|
+
const scoring = {
|
|
113
|
+
total_score: scoreDefinition(scoringRecord.total_score, "total_score"),
|
|
114
|
+
...(scoringRecord.process_score === undefined ? {} : { process_score: scoreDefinition(scoringRecord.process_score, "process_score") }),
|
|
115
|
+
};
|
|
116
|
+
if (!Array.isArray(record.tasks) || record.tasks.length === 0)
|
|
117
|
+
throw invalidInput("benchmark adapter tasks must be a non-empty array");
|
|
118
|
+
const tasks = record.tasks.map((raw, index) => {
|
|
119
|
+
const task = object(raw, `task ${index}`);
|
|
120
|
+
exact(task, ["task_id", "task_digest"], `task ${index}`);
|
|
121
|
+
return { task_id: taskId(task.task_id), task_digest: digest(task.task_digest, `task ${index} digest`) };
|
|
122
|
+
});
|
|
123
|
+
const ids = tasks.map((task) => task.task_id);
|
|
124
|
+
if (new Set(ids).size !== ids.length || JSON.stringify(ids) !== JSON.stringify([...ids].sort())) {
|
|
125
|
+
throw invalidInput("benchmark adapter task IDs must be unique and sorted");
|
|
126
|
+
}
|
|
127
|
+
return {
|
|
128
|
+
schema_version: "1",
|
|
129
|
+
kind: "gear-harbor-benchmark",
|
|
130
|
+
benchmark,
|
|
131
|
+
adapter,
|
|
132
|
+
scoring,
|
|
133
|
+
tasks,
|
|
134
|
+
dataset_digest: digest(record.dataset_digest, "dataset digest"),
|
|
135
|
+
};
|
|
136
|
+
}
|
|
137
|
+
function namedRevision(value, label) {
|
|
138
|
+
const record = object(value, label);
|
|
139
|
+
exact(record, ["id", "revision"], label);
|
|
140
|
+
return { id: identifier(record.id, `${label} id`), revision: immutableRevision(record.revision, `${label} revision`) };
|
|
141
|
+
}
|
|
142
|
+
function scoreDefinition(value, label) {
|
|
143
|
+
const record = object(value, label);
|
|
144
|
+
exact(record, ["source_metric", "direction", "range", "reducer"], label);
|
|
145
|
+
if (record.direction !== "maximize" && record.direction !== "minimize")
|
|
146
|
+
throw invalidInput(`${label} direction is invalid`);
|
|
147
|
+
if (record.reducer !== "task-macro-mean")
|
|
148
|
+
throw invalidInput(`${label} reducer is invalid`);
|
|
149
|
+
if (!Array.isArray(record.range) || record.range.length !== 2 || record.range.some((item) => typeof item !== "number" || !Number.isFinite(item))) {
|
|
150
|
+
throw invalidInput(`${label} range is invalid`);
|
|
151
|
+
}
|
|
152
|
+
const range = [record.range[0], record.range[1]];
|
|
153
|
+
if (range[0] > range[1])
|
|
154
|
+
throw invalidInput(`${label} range is reversed`);
|
|
155
|
+
return { source_metric: identifier(record.source_metric, `${label} source metric`), direction: record.direction, range, reducer: "task-macro-mean" };
|
|
156
|
+
}
|
|
157
|
+
async function taskTreeDigest(root) {
|
|
158
|
+
const rows = [];
|
|
159
|
+
const visit = async (directory) => {
|
|
160
|
+
for (const entry of (await readdir(directory, { withFileTypes: true })).sort((left, right) => left.name.localeCompare(right.name))) {
|
|
161
|
+
const absolute = path.join(directory, entry.name);
|
|
162
|
+
const relative = path.relative(root, absolute).split(path.sep).join("/");
|
|
163
|
+
const info = await lstat(absolute);
|
|
164
|
+
if (info.isSymbolicLink())
|
|
165
|
+
throw invalidInput(`benchmark task contains a symlink: ${relative}`);
|
|
166
|
+
if (info.isDirectory())
|
|
167
|
+
await visit(absolute);
|
|
168
|
+
else if (info.isFile())
|
|
169
|
+
rows.push({ path: relative, mode: info.mode & 0o111 ? "executable" : "file", sha256: createHash("sha256").update(await readFile(absolute)).digest("hex") });
|
|
170
|
+
else
|
|
171
|
+
throw invalidInput(`benchmark task contains a special file: ${relative}`);
|
|
172
|
+
}
|
|
173
|
+
};
|
|
174
|
+
await visit(root);
|
|
175
|
+
return sha256(JSON.stringify(rows));
|
|
176
|
+
}
|
|
177
|
+
function canonicalJson(value) {
|
|
178
|
+
if (Array.isArray(value))
|
|
179
|
+
return `[${value.map(canonicalJson).join(",")}]`;
|
|
180
|
+
if (value && typeof value === "object") {
|
|
181
|
+
return `{${Object.entries(value).sort(([left], [right]) => left.localeCompare(right)).map(([key, entry]) => `${JSON.stringify(key)}:${canonicalJson(entry)}`).join(",")}}`;
|
|
182
|
+
}
|
|
183
|
+
return JSON.stringify(value);
|
|
184
|
+
}
|
|
185
|
+
function sha256(value) {
|
|
186
|
+
return `sha256:${createHash("sha256").update(value).digest("hex")}`;
|
|
187
|
+
}
|
|
188
|
+
function object(value, label) {
|
|
189
|
+
if (!value || typeof value !== "object" || Array.isArray(value))
|
|
190
|
+
throw invalidInput(`${label} must be an object`);
|
|
191
|
+
return value;
|
|
192
|
+
}
|
|
193
|
+
function exact(record, fields, label) {
|
|
194
|
+
const allowed = new Set(fields);
|
|
195
|
+
const extra = Object.keys(record).find((field) => !allowed.has(field));
|
|
196
|
+
if (extra)
|
|
197
|
+
throw invalidInput(`${label} has unknown field: ${extra}`);
|
|
198
|
+
}
|
|
199
|
+
function identifier(value, label) {
|
|
200
|
+
if (typeof value !== "string" || !/^[A-Za-z0-9][A-Za-z0-9._:-]{0,255}$/.test(value))
|
|
201
|
+
throw invalidInput(`${label} is invalid`);
|
|
202
|
+
return value;
|
|
203
|
+
}
|
|
204
|
+
function immutableRevision(value, label) {
|
|
205
|
+
if (typeof value !== "string" || value.length === 0 || value.length > 512 || value.toLowerCase() === "latest")
|
|
206
|
+
throw invalidInput(`${label} is invalid`);
|
|
207
|
+
return value;
|
|
208
|
+
}
|
|
209
|
+
function taskId(value) {
|
|
210
|
+
if (typeof value !== "string" || !/^[A-Za-z0-9][A-Za-z0-9._-]{0,255}$/.test(value) || value === "." || value === "..")
|
|
211
|
+
throw invalidInput("task id is invalid");
|
|
212
|
+
return value;
|
|
213
|
+
}
|
|
214
|
+
function digest(value, label) {
|
|
215
|
+
if (typeof value !== "string" || !SHA256.test(value))
|
|
216
|
+
throw invalidInput(`${label} is invalid`);
|
|
217
|
+
return value;
|
|
218
|
+
}
|
|
219
|
+
//# sourceMappingURL=benchmark-adapter-manifest.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"benchmark-adapter-manifest.js","sourceRoot":"","sources":["../../../src/evals/benchmark-adapter-manifest.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,KAAK,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,kBAAkB,CAAC;AAC5D,OAAO,IAAI,MAAM,WAAW,CAAC;AAE7B,OAAO,EAAE,YAAY,EAAE,MAAM,wBAAwB,CAAC;AAEtD,MAAM,MAAM,GAAG,uBAAuB,CAAC;AAwBvC,+EAA+E;AAC/E,MAAM,CAAC,KAAK,UAAU,6BAA6B,CAAC,KAMnD;IACC,MAAM,GAAG,GAAG,CAAC,GAAG,KAAK,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,CAAC;IACtC,IAAI,CAAC,GAAG,CAAC,MAAM,IAAI,IAAI,GAAG,CAAC,GAAG,CAAC,CAAC,IAAI,KAAK,GAAG,CAAC,MAAM;QAAE,MAAM,YAAY,CAAC,yDAAyD,CAAC,CAAC;IACnI,MAAM,KAAK,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,KAAK,EAAE,EAAE,EAAE,EAAE;QACnD,MAAM,KAAK,GAAG,MAAM,CAAC,EAAE,CAAC,CAAC;QACzB,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,OAAO,CAAC,EAAE,KAAK,CAAC,CAAC;QAChE,MAAM,IAAI,GAAG,MAAM,KAAK,CAAC,SAAS,CAAC,CAAC,KAAK,CAAC,CAAC,KAA4B,EAAE,EAAE;YACzE,IAAI,KAAK,CAAC,IAAI,KAAK,QAAQ;gBAAE,MAAM,YAAY,CAAC,6BAA6B,KAAK,EAAE,CAAC,CAAC;YACtF,MAAM,KAAK,CAAC;QACd,CAAC,CAAC,CAAC;QACH,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,IAAI,IAAI,CAAC,cAAc,EAAE;YAAE,MAAM,YAAY,CAAC,qCAAqC,KAAK,EAAE,CAAC,CAAC;QACnH,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,WAAW,EAAE,MAAM,cAAc,CAAC,SAAS,CAAC,EAAE,CAAC;IAC1E,CAAC,CAAC,CAAC,CAAC;IACJ,MAAM,IAAI,GAAG;QACX,cAAc,EAAE,GAAY;QAC5B,IAAI,EAAE,uBAAgC;QACtC,SAAS,EAAE,KAAK,CAAC,SAAS;QAC1B,OAAO,EAAE,KAAK,CAAC,OAAO;QACtB,OAAO,EAAE,KAAK,CAAC,OAAO;QACtB,KAAK;KACN,CAAC;IACF,OAAO,EAAE,GAAG,IAAI,EAAE,cAAc,EAAE,MAAM,CAAC,aAAa,CAAC,IAAI,CAAC,CAAC,EAAE,CAAC;AAClE,CAAC;AAED,gFAAgF;AAChF,MAAM,CAAC,KAAK,UAAU,4BAA4B,CAAC,OAAe;IAChE,MAAM,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;IACnC,IAAI,GAAY,CAAC;IACjB,IAAI,CAAC;QACH,MAAM,IAAI,GAAG,MAAM,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,wBAAwB,CAAC,CAAC,CAAC;QACpE,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE,IAAI,IAAI,CAAC,cAAc,EAAE;YAAE,MAAM,YAAY,CAAC,+CAA+C,CAAC,CAAC;QACjH,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,wBAAwB,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC;IACtF,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAK,KAA+B,CAAC,IAAI,KAAK,QAAQ;YAAE,OAAO,IAAI,CAAC;QACpE,IAAI,KAAK,YAAY,WAAW;YAAE,MAAM,YAAY,CAAC,wCAAwC,CAAC,CAAC;QAC/F,MAAM,KAAK,CAAC;IACd,CAAC;IACD,MAAM,QAAQ,GAAG,aAAa,CAAC,GAAG,CAAC,CAAC;IACpC,MAAM,WAAW,GAAoD,EAAE,CAAC;IACxE,KAAK,MAAM,IAAI,IAAI,QAAQ,CAAC,KAAK,EAAE,CAAC;QAClC,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,IAAI,CAAC,OAAO,CAAC,CAAC;QAChD,MAAM,IAAI,GAAG,MAAM,KAAK,CAAC,SAAS,CAAC,CAAC,KAAK,CAAC,CAAC,KAA4B,EAAE,EAAE;YACzE,IAAI,KAAK,CAAC,IAAI,KAAK,QAAQ;gBAAE,MAAM,YAAY,CAAC,6BAA6B,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC;YAC7F,MAAM,KAAK,CAAC;QACd,CAAC,CAAC,CAAC;QACH,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,IAAI,IAAI,CAAC,cAAc,EAAE;YAAE,MAAM,YAAY,CAAC,qCAAqC,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC;QAC1H,MAAM,MAAM,GAAG,MAAM,cAAc,CAAC,SAAS,CAAC,CAAC;QAC/C,IAAI,MAAM,KAAK,IAAI,CAAC,WAAW;YAAE,MAAM,YAAY,CAAC,kCAAkC,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC;QACtG,WAAW,CAAC,IAAI,CAAC,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,WAAW,EAAE,MAAM,EAAE,CAAC,CAAC;IACnE,CAAC;IACD,MAAM,eAAe,GAAa,EAAE,CAAC;IACrC,KAAK,MAAM,KAAK,IAAI,MAAM,OAAO,CAAC,IAAI,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,EAAE,CAAC;QACjE,IAAI,CAAC,KAAK,CAAC,WAAW,EAAE;YAAE,SAAS;QACnC,IAAI,CAAC;YACH,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,WAAW,CAAC,CAAC,CAAC;YACvE,IAAI,QAAQ,CAAC,MAAM,EAAE,IAAI,CAAC,QAAQ,CAAC,cAAc,EAAE;gBAAE,eAAe,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;QACxF,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,IAAK,KAA+B,CAAC,IAAI,KAAK,QAAQ;gBAAE,MAAM,KAAK,CAAC;QACtE,CAAC;IACH,CAAC;IACD,IAAI,IAAI,CAAC,SAAS,CAAC,eAAe,CAAC,IAAI,EAAE,CAAC,KAAK,IAAI,CAAC,SAAS,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,EAAE,CAAC;QACvG,MAAM,YAAY,CAAC,gEAAgE,CAAC,CAAC;IACvF,CAAC;IACD,MAAM,IAAI,GAAG;QACX,cAAc,EAAE,QAAQ,CAAC,cAAc;QACvC,IAAI,EAAE,QAAQ,CAAC,IAAI;QACnB,SAAS,EAAE,QAAQ,CAAC,SAAS;QAC7B,OAAO,EAAE,QAAQ,CAAC,OAAO;QACzB,OAAO,EAAE,QAAQ,CAAC,OAAO;QACzB,KAAK,EAAE,WAAW;KACnB,CAAC;IACF,MAAM,QAAQ,GAAG,MAAM,CAAC,aAAa,CAAC,IAAI,CAAC,CAAC,CAAC;IAC7C,IAAI,QAAQ,CAAC,cAAc,KAAK,QAAQ;QAAE,MAAM,YAAY,CAAC,2CAA2C,CAAC,CAAC;IAC1G,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,MAAM,UAAU,gBAAgB,CAAC,KAAa,EAAE,UAAsC;IACpF,OAAO,KAAK,IAAI,UAAU,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,KAAK,IAAI,UAAU,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;AACtE,CAAC;AAED,SAAS,aAAa,CAAC,KAAc;IACnC,MAAM,MAAM,GAAG,MAAM,CAAC,KAAK,EAAE,4BAA4B,CAAC,CAAC;IAC3D,KAAK,CAAC,MAAM,EAAE,CAAC,gBAAgB,EAAE,MAAM,EAAE,WAAW,EAAE,SAAS,EAAE,SAAS,EAAE,OAAO,EAAE,gBAAgB,CAAC,EAAE,4BAA4B,CAAC,CAAC;IACtI,IAAI,MAAM,CAAC,cAAc,KAAK,GAAG,IAAI,MAAM,CAAC,IAAI,KAAK,uBAAuB;QAAE,MAAM,YAAY,CAAC,wCAAwC,CAAC,CAAC;IAC3I,MAAM,SAAS,GAAG,aAAa,CAAC,MAAM,CAAC,SAAS,EAAE,WAAW,CAAC,CAAC;IAC/D,MAAM,aAAa,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC;IACxD,KAAK,CAAC,aAAa,EAAE,CAAC,IAAI,EAAE,UAAU,EAAE,iBAAiB,CAAC,EAAE,SAAS,CAAC,CAAC;IACvE,IAAI,aAAa,CAAC,eAAe,KAAK,4BAA4B;QAAE,MAAM,YAAY,CAAC,+CAA+C,CAAC,CAAC;IACxI,MAAM,OAAO,GAAG,EAAE,EAAE,EAAE,UAAU,CAAC,aAAa,CAAC,EAAE,EAAE,YAAY,CAAC,EAAE,QAAQ,EAAE,iBAAiB,CAAC,aAAa,CAAC,QAAQ,EAAE,kBAAkB,CAAC,EAAE,eAAe,EAAE,aAAa,CAAC,eAAe,EAAW,CAAC;IACrM,MAAM,aAAa,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC;IACxD,KAAK,CAAC,aAAa,EAAE,CAAC,aAAa,EAAE,eAAe,CAAC,EAAE,SAAS,CAAC,CAAC;IAClE,MAAM,OAAO,GAA6B;QACxC,WAAW,EAAE,eAAe,CAAC,aAAa,CAAC,WAAW,EAAE,aAAa,CAAC;QACtE,GAAG,CAAC,aAAa,CAAC,aAAa,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,aAAa,EAAE,eAAe,CAAC,aAAa,CAAC,aAAa,EAAE,eAAe,CAAC,EAAE,CAAC;KACvI,CAAC;IACF,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,MAAM,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,MAAM,YAAY,CAAC,mDAAmD,CAAC,CAAC;IACvI,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE;QAC5C,MAAM,IAAI,GAAG,MAAM,CAAC,GAAG,EAAE,QAAQ,KAAK,EAAE,CAAC,CAAC;QAC1C,KAAK,CAAC,IAAI,EAAE,CAAC,SAAS,EAAE,aAAa,CAAC,EAAE,QAAQ,KAAK,EAAE,CAAC,CAAC;QACzD,OAAO,EAAE,OAAO,EAAE,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,EAAE,WAAW,EAAE,MAAM,CAAC,IAAI,CAAC,WAAW,EAAE,QAAQ,KAAK,SAAS,CAAC,EAAE,CAAC;IAC1G,CAAC,CAAC,CAAC;IACH,MAAM,GAAG,GAAG,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;IAC9C,IAAI,IAAI,GAAG,CAAC,GAAG,CAAC,CAAC,IAAI,KAAK,GAAG,CAAC,MAAM,IAAI,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,KAAK,IAAI,CAAC,SAAS,CAAC,CAAC,GAAG,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC,EAAE,CAAC;QAChG,MAAM,YAAY,CAAC,sDAAsD,CAAC,CAAC;IAC7E,CAAC;IACD,OAAO;QACL,cAAc,EAAE,GAAG;QACnB,IAAI,EAAE,uBAAuB;QAC7B,SAAS;QACT,OAAO;QACP,OAAO;QACP,KAAK;QACL,cAAc,EAAE,MAAM,CAAC,MAAM,CAAC,cAAc,EAAE,gBAAgB,CAAC;KAChE,CAAC;AACJ,CAAC;AAED,SAAS,aAAa,CAAC,KAAc,EAAE,KAAa;IAClD,MAAM,MAAM,GAAG,MAAM,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC;IACpC,KAAK,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,UAAU,CAAC,EAAE,KAAK,CAAC,CAAC;IACzC,OAAO,EAAE,EAAE,EAAE,UAAU,CAAC,MAAM,CAAC,EAAE,EAAE,GAAG,KAAK,KAAK,CAAC,EAAE,QAAQ,EAAE,iBAAiB,CAAC,MAAM,CAAC,QAAQ,EAAE,GAAG,KAAK,WAAW,CAAC,EAAE,CAAC;AACzH,CAAC;AAED,SAAS,eAAe,CAAC,KAAc,EAAE,KAAa;IACpD,MAAM,MAAM,GAAG,MAAM,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC;IACpC,KAAK,CAAC,MAAM,EAAE,CAAC,eAAe,EAAE,WAAW,EAAE,OAAO,EAAE,SAAS,CAAC,EAAE,KAAK,CAAC,CAAC;IACzE,IAAI,MAAM,CAAC,SAAS,KAAK,UAAU,IAAI,MAAM,CAAC,SAAS,KAAK,UAAU;QAAE,MAAM,YAAY,CAAC,GAAG,KAAK,uBAAuB,CAAC,CAAC;IAC5H,IAAI,MAAM,CAAC,OAAO,KAAK,iBAAiB;QAAE,MAAM,YAAY,CAAC,GAAG,KAAK,qBAAqB,CAAC,CAAC;IAC5F,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,MAAM,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,IAAI,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,OAAO,IAAI,KAAK,QAAQ,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,EAAE,CAAC;QACjJ,MAAM,YAAY,CAAC,GAAG,KAAK,mBAAmB,CAAC,CAAC;IAClD,CAAC;IACD,MAAM,KAAK,GAAG,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAW,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC,CAAW,CAAU,CAAC;IAC9E,IAAI,KAAK,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC,CAAC,CAAC;QAAE,MAAM,YAAY,CAAC,GAAG,KAAK,oBAAoB,CAAC,CAAC;IAC1E,OAAO,EAAE,aAAa,EAAE,UAAU,CAAC,MAAM,CAAC,aAAa,EAAE,GAAG,KAAK,gBAAgB,CAAC,EAAE,SAAS,EAAE,MAAM,CAAC,SAAS,EAAE,KAAK,EAAE,OAAO,EAAE,iBAAiB,EAAE,CAAC;AACvJ,CAAC;AAED,KAAK,UAAU,cAAc,CAAC,IAAY;IACxC,MAAM,IAAI,GAAyE,EAAE,CAAC;IACtF,MAAM,KAAK,GAAG,KAAK,EAAE,SAAiB,EAAiB,EAAE;QACvD,KAAK,MAAM,KAAK,IAAI,CAAC,MAAM,OAAO,CAAC,SAAS,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,EAAE,CAAC;YACnI,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC;YAClD,MAAM,QAAQ,GAAG,IAAI,CAAC,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;YACzE,MAAM,IAAI,GAAG,MAAM,KAAK,CAAC,QAAQ,CAAC,CAAC;YACnC,IAAI,IAAI,CAAC,cAAc,EAAE;gBAAE,MAAM,YAAY,CAAC,sCAAsC,QAAQ,EAAE,CAAC,CAAC;YAChG,IAAI,IAAI,CAAC,WAAW,EAAE;gBAAE,MAAM,KAAK,CAAC,QAAQ,CAAC,CAAC;iBACzC,IAAI,IAAI,CAAC,MAAM,EAAE;gBAAE,IAAI,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,GAAG,KAAK,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,MAAM,EAAE,MAAM,EAAE,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,MAAM,QAAQ,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;;gBAC/K,MAAM,YAAY,CAAC,2CAA2C,QAAQ,EAAE,CAAC,CAAC;QACjF,CAAC;IACH,CAAC,CAAC;IACF,MAAM,KAAK,CAAC,IAAI,CAAC,CAAC;IAClB,OAAO,MAAM,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC;AACtC,CAAC;AAED,SAAS,aAAa,CAAC,KAAc;IACnC,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;QAAE,OAAO,IAAI,KAAK,CAAC,GAAG,CAAC,aAAa,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC;IAC3E,IAAI,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;QACvC,OAAO,IAAI,MAAM,CAAC,OAAO,CAAC,KAAgC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,KAAK,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,IAAI,aAAa,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC;IACxM,CAAC;IACD,OAAO,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;AAC/B,CAAC;AAED,SAAS,MAAM,CAAC,KAAa;IAC3B,OAAO,UAAU,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAY,CAAC;AAChF,CAAC;AAED,SAAS,MAAM,CAAC,KAAc,EAAE,KAAa;IAC3C,IAAI,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;QAAE,MAAM,YAAY,CAAC,GAAG,KAAK,oBAAoB,CAAC,CAAC;IAClH,OAAO,KAAgC,CAAC;AAC1C,CAAC;AAED,SAAS,KAAK,CAAC,MAA+B,EAAE,MAAyB,EAAE,KAAa;IACtF,MAAM,OAAO,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,CAAC;IAChC,MAAM,KAAK,GAAG,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC;IACvE,IAAI,KAAK;QAAE,MAAM,YAAY,CAAC,GAAG,KAAK,uBAAuB,KAAK,EAAE,CAAC,CAAC;AACxE,CAAC;AAED,SAAS,UAAU,CAAC,KAAc,EAAE,KAAa;IAC/C,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,CAAC,qCAAqC,CAAC,IAAI,CAAC,KAAK,CAAC;QAAE,MAAM,YAAY,CAAC,GAAG,KAAK,aAAa,CAAC,CAAC;IAC/H,OAAO,KAAK,CAAC;AACf,CAAC;AAED,SAAS,iBAAiB,CAAC,KAAc,EAAE,KAAa;IACtD,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,IAAI,KAAK,CAAC,MAAM,GAAG,GAAG,IAAI,KAAK,CAAC,WAAW,EAAE,KAAK,QAAQ;QAAE,MAAM,YAAY,CAAC,GAAG,KAAK,aAAa,CAAC,CAAC;IACzJ,OAAO,KAAK,CAAC;AACf,CAAC;AAED,SAAS,MAAM,CAAC,KAAc;IAC5B,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,CAAC,oCAAoC,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,KAAK,KAAK,GAAG,IAAI,KAAK,KAAK,IAAI;QAAE,MAAM,YAAY,CAAC,oBAAoB,CAAC,CAAC;IAChK,OAAO,KAAK,CAAC;AACf,CAAC;AAED,SAAS,MAAM,CAAC,KAAc,EAAE,KAAa;IAC3C,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC;QAAE,MAAM,YAAY,CAAC,GAAG,KAAK,aAAa,CAAC,CAAC;IAChG,OAAO,KAAe,CAAC;AACzB,CAAC"}
|
|
@@ -0,0 +1,36 @@
|
|
|
1
|
+
import { readFile } from "node:fs/promises";
|
|
2
|
+
import path from "node:path";
|
|
3
|
+
import { invalidInput } from "../foundation/index.js";
|
|
4
|
+
export function assertBenchmarkCandidate(tasks, harnessId, agentArgs) {
|
|
5
|
+
if (tasks.some(task => task.driver.kind === "model-call") && (harnessId !== "model-call" || agentArgs.length)) {
|
|
6
|
+
throw invalidInput("no-tools tasks require the trusted model-call harness without agent overrides");
|
|
7
|
+
}
|
|
8
|
+
if (tasks.some(task => task.driver.kind !== "model-call" && task.requirements.includes("native-image-input")) && harnessId !== "codex") {
|
|
9
|
+
throw invalidInput("native-image agent tasks currently require the Codex image-capable harness");
|
|
10
|
+
}
|
|
11
|
+
}
|
|
12
|
+
/** The verified task trees contain the Package v1 requirements after export. */
|
|
13
|
+
export async function assertStandardBenchmarkCandidate(dataset, manifest, harnessId, agentArgs) {
|
|
14
|
+
const tasks = [];
|
|
15
|
+
for (const { task_id } of manifest.tasks) {
|
|
16
|
+
let descriptor;
|
|
17
|
+
try {
|
|
18
|
+
descriptor = JSON.parse(await readFile(path.join(dataset, task_id, ".hitch-benchmark.json"), "utf8"));
|
|
19
|
+
}
|
|
20
|
+
catch (error) {
|
|
21
|
+
// Direct Harbor adapters need not use the Package v1 execution bridge.
|
|
22
|
+
if (error.code === "ENOENT")
|
|
23
|
+
continue;
|
|
24
|
+
throw error;
|
|
25
|
+
}
|
|
26
|
+
const task = descriptor?.task;
|
|
27
|
+
if (descriptor?.schema_version !== "1" || descriptor.task_id !== task_id
|
|
28
|
+
|| !["model-call", "terminal", "tool-server"].includes(task?.driver?.kind)
|
|
29
|
+
|| !Array.isArray(task?.requirements) || task.requirements.some((item) => typeof item !== "string")) {
|
|
30
|
+
throw invalidInput(`invalid compiled candidate requirements: ${task_id}`);
|
|
31
|
+
}
|
|
32
|
+
tasks.push(task);
|
|
33
|
+
}
|
|
34
|
+
assertBenchmarkCandidate(tasks, harnessId, agentArgs);
|
|
35
|
+
}
|
|
36
|
+
//# sourceMappingURL=benchmark-candidate.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"benchmark-candidate.js","sourceRoot":"","sources":["../../../src/evals/benchmark-candidate.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAC;AAC5C,OAAO,IAAI,MAAM,WAAW,CAAC;AAC7B,OAAO,EAAE,YAAY,EAAE,MAAM,wBAAwB,CAAC;AAQtD,MAAM,UAAU,wBAAwB,CAAC,KAAuC,EAAE,SAAiB,EAAE,SAA4B;IAC/H,IAAI,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,MAAM,CAAC,IAAI,KAAK,YAAY,CAAC,IAAI,CAAC,SAAS,KAAK,YAAY,IAAI,SAAS,CAAC,MAAM,CAAC,EAAE,CAAC;QAC9G,MAAM,YAAY,CAAC,+EAA+E,CAAC,CAAC;IACtG,CAAC;IACD,IAAI,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,MAAM,CAAC,IAAI,KAAK,YAAY,IAAI,IAAI,CAAC,YAAY,CAAC,QAAQ,CAAC,oBAAoB,CAAC,CAAC,IAAI,SAAS,KAAK,OAAO,EAAE,CAAC;QACvI,MAAM,YAAY,CAAC,4EAA4E,CAAC,CAAC;IACnG,CAAC;AACH,CAAC;AAED,gFAAgF;AAChF,MAAM,CAAC,KAAK,UAAU,gCAAgC,CACpD,OAAe,EACf,QAAoC,EACpC,SAAiB,EACjB,SAA4B;IAE5B,MAAM,KAAK,GAA4B,EAAE,CAAC;IAC1C,KAAK,MAAM,EAAE,OAAO,EAAE,IAAI,QAAQ,CAAC,KAAK,EAAE,CAAC;QACzC,IAAI,UAAU,CAAC;QACf,IAAI,CAAC;YACH,UAAU,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,uBAAuB,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC;QACxG,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,uEAAuE;YACvE,IAAK,KAA+B,CAAC,IAAI,KAAK,QAAQ;gBAAE,SAAS;YACjE,MAAM,KAAK,CAAC;QACd,CAAC;QACD,MAAM,IAAI,GAAG,UAAU,EAAE,IAAI,CAAC;QAC9B,IAAI,UAAU,EAAE,cAAc,KAAK,GAAG,IAAI,UAAU,CAAC,OAAO,KAAK,OAAO;eACnE,CAAC,CAAC,YAAY,EAAE,UAAU,EAAE,aAAa,CAAC,CAAC,QAAQ,CAAC,IAAI,EAAE,MAAM,EAAE,IAAI,CAAC;eACvE,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,EAAE,YAAY,CAAC,IAAI,IAAI,CAAC,YAAY,CAAC,IAAI,CAAC,CAAC,IAAa,EAAE,EAAE,CAAC,OAAO,IAAI,KAAK,QAAQ,CAAC,EAAE,CAAC;YAC/G,MAAM,YAAY,CAAC,4CAA4C,OAAO,EAAE,CAAC,CAAC;QAC5E,CAAC;QACD,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACnB,CAAC;IACD,wBAAwB,CAAC,KAAK,EAAE,SAAS,EAAE,SAAS,CAAC,CAAC;AACxD,CAAC"}
|
|
@@ -0,0 +1,166 @@
|
|
|
1
|
+
import { cp, mkdir, mkdtemp, readFile, readdir, rename, rm, writeFile } from "node:fs/promises";
|
|
2
|
+
import path from "node:path";
|
|
3
|
+
import { stringify as stringifyTOML } from "smol-toml";
|
|
4
|
+
import { benchmarkTreeDigest, loadBenchmark, loadBenchmarkLock } from "../benchmarks/index.js";
|
|
5
|
+
import { atomicWriteJSON, ensureDir, invalidInput, sha256JSON, statePaths, withFileLock } from "../foundation/index.js";
|
|
6
|
+
import { newEvalId, validateEvalRequest } from "./request.js";
|
|
7
|
+
import { parseHarnessReference } from "../revisions/index.js";
|
|
8
|
+
import { assertBenchmarkCandidate } from "./benchmark-candidate.js";
|
|
9
|
+
import { buildBenchmarkAdapterManifest, loadBenchmarkAdapterManifest } from "./benchmark-adapter-manifest.js";
|
|
10
|
+
import { runEval } from "./service.js";
|
|
11
|
+
export const STANDARD_BENCHMARK_COMPILER = "harbor-package@6";
|
|
12
|
+
export async function runBenchmarkEval(options) {
|
|
13
|
+
if (Boolean(options.benchmark) === Boolean(options.benchmarkLock))
|
|
14
|
+
throw invalidInput("provide exactly one of --benchmark or --benchmark-lock");
|
|
15
|
+
if (options.request.dataset)
|
|
16
|
+
throw invalidInput("--dataset and --benchmark are mutually exclusive");
|
|
17
|
+
if (options.resumeExisting || options.precreated || options.remoteWorkExecutor || options.normalizedRequest)
|
|
18
|
+
throw invalidInput("standard benchmark packages currently support local managed runs only");
|
|
19
|
+
const loaded = options.benchmark ? await loadBenchmark(options.benchmark) : await loadBenchmarkLock(options.benchmarkLock);
|
|
20
|
+
assertBenchmarkCandidate(loaded.tasks.map(task => task.config), parseHarnessReference(String(options.request.harness_ref)).harness_id, Array.isArray(options.request.agent_args) ? options.request.agent_args : []);
|
|
21
|
+
const evalId = options.evalId ?? newEvalId();
|
|
22
|
+
const compiled = await compileBenchmark(loaded, options.root);
|
|
23
|
+
const request = {
|
|
24
|
+
...options.request, dataset: compiled.tasks,
|
|
25
|
+
attempts: options.request.attempts ?? loaded.profile.sampling.attempts_per_task,
|
|
26
|
+
max_concurrent: options.request.max_concurrent ?? 1,
|
|
27
|
+
// Zero disables the legacy blanket timeout; each task/bridge owns its budget.
|
|
28
|
+
timeout_ms: options.request.timeout_ms ?? 0,
|
|
29
|
+
setup_timeout_ms: options.request.setup_timeout_ms ?? loaded.profile.budget.setup_timeout_ms,
|
|
30
|
+
infrastructure_retries: options.request.infrastructure_retries ?? loaded.profile.grading.infrastructure_retries,
|
|
31
|
+
};
|
|
32
|
+
const directory = path.join(statePaths(options.root).evals, evalId, "benchmark");
|
|
33
|
+
const normalizedRequest = await validateEvalRequest(request);
|
|
34
|
+
if (normalizedRequest.benchmark_id !== loaded.manifest.id)
|
|
35
|
+
throw invalidInput("compiled benchmark manifest identity mismatch");
|
|
36
|
+
return runEval({ ...options, evalId, request, normalizedRequest, executionStrategy: "local-task-slots-v1", onControlPhase: async (phase, work) => {
|
|
37
|
+
if (phase === "planning") {
|
|
38
|
+
await ensureDir(directory);
|
|
39
|
+
await atomicWriteJSON(path.join(directory, "benchmark.lock.json"), loaded.lock);
|
|
40
|
+
await atomicWriteJSON(path.join(directory, "manifest.json"), loaded.manifest);
|
|
41
|
+
await atomicWriteJSON(path.join(directory, "effective-profile.json"), {
|
|
42
|
+
...loaded.profile,
|
|
43
|
+
effective: { attempts: request.attempts, max_concurrent: request.max_concurrent, timeout_ms: request.timeout_ms, seed_applied_to_model: false },
|
|
44
|
+
});
|
|
45
|
+
await atomicWriteJSON(path.join(directory, "package.json"), { source: compiled.source, tasks: compiled.tasks, package_digest: loaded.lock.package_digest, compiled_digest: compiled.digest });
|
|
46
|
+
}
|
|
47
|
+
await options.onControlPhase?.(phase, work);
|
|
48
|
+
} });
|
|
49
|
+
}
|
|
50
|
+
export async function compileBenchmark(loaded, root) {
|
|
51
|
+
const digest = sha256JSON({ lock: loaded.lock, compiler: STANDARD_BENCHMARK_COMPILER });
|
|
52
|
+
const store = await ensureDir(path.join(root, "store", "benchmarks"));
|
|
53
|
+
const target = path.join(store, digest.slice(7));
|
|
54
|
+
await withFileLock(path.join(store, "locks"), digest, async () => {
|
|
55
|
+
try {
|
|
56
|
+
const existing = await loadBenchmark(path.join(target, "source"));
|
|
57
|
+
if (sha256JSON(existing.lock) !== sha256JSON(loaded.lock))
|
|
58
|
+
throw invalidInput("stored benchmark integrity mismatch");
|
|
59
|
+
const metadata = JSON.parse(await readFile(path.join(target, "compiled.json"), "utf8"));
|
|
60
|
+
if (metadata.digest !== digest)
|
|
61
|
+
throw invalidInput("stored benchmark compiler identity mismatch");
|
|
62
|
+
if (metadata.tasks_digest !== await benchmarkTreeDigest(path.join(target, "tasks")))
|
|
63
|
+
throw invalidInput("compiled benchmark tasks were modified");
|
|
64
|
+
return;
|
|
65
|
+
}
|
|
66
|
+
catch (error) {
|
|
67
|
+
if (error.code !== "ENOENT")
|
|
68
|
+
throw error;
|
|
69
|
+
}
|
|
70
|
+
const temporary = await mkdtemp(path.join(store, ".building-"));
|
|
71
|
+
try {
|
|
72
|
+
await cp(loaded.directory, path.join(temporary, "source"), { recursive: true, dereference: false });
|
|
73
|
+
const snapshot = await loadBenchmark(path.join(temporary, "source"));
|
|
74
|
+
if (sha256JSON(snapshot.lock) !== sha256JSON(loaded.lock))
|
|
75
|
+
throw invalidInput("benchmark changed while being sealed");
|
|
76
|
+
await mkdir(path.join(temporary, "tasks"));
|
|
77
|
+
for (const task of snapshot.tasks) {
|
|
78
|
+
const taskPath = path.join(temporary, "tasks", task.id);
|
|
79
|
+
await cp(path.join(snapshot.directory, task.path), taskPath, { recursive: true });
|
|
80
|
+
const nativePhases = task.config.driver.kind === "tool-server" && task.config.driver.config.native_phases;
|
|
81
|
+
let finalizationTimeoutMs = loaded.profile.budget.collection_timeout_ms + loaded.profile.budget.cleanup_grace_ms;
|
|
82
|
+
if (nativePhases) {
|
|
83
|
+
// Hitch enforces the original candidate budget in its private phase
|
|
84
|
+
// supervisor. Harbor's outer guard also allows final evidence export
|
|
85
|
+
// and failure teardown without extending the model's deadline.
|
|
86
|
+
// Cancellation/export and host inspection each have a shutdown
|
|
87
|
+
// allowance, bounded by collection_timeout_ms; snapshot has its own.
|
|
88
|
+
const collections = nativePhases.protocol === "hitch-native-phase-control@2" ? 4 : 3;
|
|
89
|
+
finalizationTimeoutMs = collections * loaded.profile.budget.collection_timeout_ms + 2 * loaded.profile.budget.cleanup_grace_ms;
|
|
90
|
+
}
|
|
91
|
+
// The inner Hitch invocation retains the source budget. The outer
|
|
92
|
+
// Harbor guard must also permit process retirement and result export.
|
|
93
|
+
const execution = structuredClone(task.harbor);
|
|
94
|
+
const agent = execution.agent;
|
|
95
|
+
agent.timeout_sec = Number(agent.timeout_sec) + Math.ceil(finalizationTimeoutMs / 1000);
|
|
96
|
+
await writeFile(path.join(taskPath, "task.toml"), stringifyTOML(execution));
|
|
97
|
+
await atomicWriteJSON(path.join(taskPath, ".hitch-benchmark.json"), {
|
|
98
|
+
schema_version: "1", task_id: task.id, task: task.config,
|
|
99
|
+
profile_digest: loaded.lock.profile_digest, profile: loaded.profile,
|
|
100
|
+
primary_metric: loaded.manifest.primary_metric, metrics: loaded.manifest.metrics,
|
|
101
|
+
score_contract: { total_score: loaded.manifest.primary_metric },
|
|
102
|
+
tools: task.tools, agent_timeout_sec: task.harbor.agent.timeout_sec,
|
|
103
|
+
agent_finalization_timeout_ms: finalizationTimeoutMs,
|
|
104
|
+
...(task.config.driver.kind === "model-call" ? { candidate_input: JSON.parse(await readFile(path.join(snapshot.directory, task.config.driver.config.input), "utf8")) } : {}),
|
|
105
|
+
package_digest: loaded.lock.package_digest, task_digest: loaded.lock.tasks.find((t) => t.task_id === task.id).task_digest,
|
|
106
|
+
});
|
|
107
|
+
}
|
|
108
|
+
const primary = loaded.manifest.metrics[loaded.manifest.primary_metric];
|
|
109
|
+
const standardManifest = await buildBenchmarkAdapterManifest({
|
|
110
|
+
dataset: path.join(temporary, "tasks"),
|
|
111
|
+
benchmark: { id: loaded.manifest.id, revision: loaded.lock.package_digest },
|
|
112
|
+
adapter: { id: "hitch-package-v1-compiler", revision: digest, output_protocol: "gear-harbor-eval-result-v1" },
|
|
113
|
+
scoring: {
|
|
114
|
+
total_score: {
|
|
115
|
+
source_metric: loaded.manifest.primary_metric,
|
|
116
|
+
direction: primary.direction,
|
|
117
|
+
range: primary.range,
|
|
118
|
+
reducer: "task-macro-mean",
|
|
119
|
+
},
|
|
120
|
+
},
|
|
121
|
+
taskIds: snapshot.tasks.map((task) => task.id),
|
|
122
|
+
});
|
|
123
|
+
await atomicWriteJSON(path.join(temporary, "tasks", "benchmark.adapter.json"), standardManifest);
|
|
124
|
+
await atomicWriteJSON(path.join(temporary, "source", "benchmark.lock.json"), loaded.lock);
|
|
125
|
+
await atomicWriteJSON(path.join(temporary, "compiled.json"), { digest, tasks_digest: await benchmarkTreeDigest(path.join(temporary, "tasks")) });
|
|
126
|
+
await rename(temporary, target);
|
|
127
|
+
}
|
|
128
|
+
finally {
|
|
129
|
+
await rm(temporary, { recursive: true, force: true });
|
|
130
|
+
}
|
|
131
|
+
});
|
|
132
|
+
return { source: path.join(target, "source"), tasks: path.join(target, "tasks"), digest };
|
|
133
|
+
}
|
|
134
|
+
/** Compile any supported Package v1 benchmark into a standalone Harbor dataset. */
|
|
135
|
+
export async function exportStandardBenchmarkDataset(packageDirectory, outputDirectory) {
|
|
136
|
+
const output = path.resolve(outputDirectory);
|
|
137
|
+
await mkdir(path.dirname(output), { recursive: true });
|
|
138
|
+
await mkdir(output); // Refuse to overwrite any existing path.
|
|
139
|
+
const temporaryRoot = await mkdtemp(path.join(path.dirname(output), ".hitch-standard-compile-"));
|
|
140
|
+
try {
|
|
141
|
+
const loaded = await loadBenchmark(packageDirectory);
|
|
142
|
+
const compiled = await compileBenchmark(loaded, temporaryRoot);
|
|
143
|
+
for (const entry of await readdir(compiled.tasks)) {
|
|
144
|
+
await cp(path.join(compiled.tasks, entry), path.join(output, entry), { recursive: true, errorOnExist: true, force: false });
|
|
145
|
+
}
|
|
146
|
+
const manifest = await loadBenchmarkAdapterManifest(output);
|
|
147
|
+
if (!manifest)
|
|
148
|
+
throw invalidInput("compiled benchmark did not produce a standard manifest");
|
|
149
|
+
return {
|
|
150
|
+
dataset: output,
|
|
151
|
+
benchmark_id: manifest.benchmark.id,
|
|
152
|
+
benchmark_revision: manifest.dataset_digest,
|
|
153
|
+
source_package_digest: loaded.lock.package_digest,
|
|
154
|
+
tasks: manifest.tasks.map((task) => task.task_id),
|
|
155
|
+
scoring: manifest.scoring,
|
|
156
|
+
};
|
|
157
|
+
}
|
|
158
|
+
catch (error) {
|
|
159
|
+
await rm(output, { recursive: true, force: true });
|
|
160
|
+
throw error;
|
|
161
|
+
}
|
|
162
|
+
finally {
|
|
163
|
+
await rm(temporaryRoot, { recursive: true, force: true });
|
|
164
|
+
}
|
|
165
|
+
}
|
|
166
|
+
//# sourceMappingURL=benchmark-run.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"benchmark-run.js","sourceRoot":"","sources":["../../../src/evals/benchmark-run.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,OAAO,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,EAAE,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAC;AAChG,OAAO,IAAI,MAAM,WAAW,CAAC;AAC7B,OAAO,EAAE,SAAS,IAAI,aAAa,EAAE,MAAM,WAAW,CAAC;AACvD,OAAO,EAAE,mBAAmB,EAAE,aAAa,EAAE,iBAAiB,EAAE,MAAM,wBAAwB,CAAC;AAE/F,OAAO,EAAE,eAAe,EAAE,SAAS,EAAE,YAAY,EAAE,UAAU,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,wBAAwB,CAAC;AACxH,OAAO,EAAE,SAAS,EAAE,mBAAmB,EAAE,MAAM,cAAc,CAAC;AAC9D,OAAO,EAAE,qBAAqB,EAAE,MAAM,uBAAuB,CAAC;AAC9D,OAAO,EAAE,wBAAwB,EAAE,MAAM,0BAA0B,CAAC;AACpE,OAAO,EAAE,6BAA6B,EAAE,4BAA4B,EAAE,MAAM,iCAAiC,CAAC;AAC9G,OAAO,EAAE,OAAO,EAAE,MAAM,cAAc,CAAC;AAGvC,MAAM,CAAC,MAAM,2BAA2B,GAAG,kBAAkB,CAAC;AAE9D,MAAM,CAAC,KAAK,UAAU,gBAAgB,CAAC,OAItC;IACC,IAAI,OAAO,CAAC,OAAO,CAAC,SAAS,CAAC,KAAK,OAAO,CAAC,OAAO,CAAC,aAAa,CAAC;QAAE,MAAM,YAAY,CAAC,wDAAwD,CAAC,CAAC;IAChJ,IAAI,OAAO,CAAC,OAAO,CAAC,OAAO;QAAE,MAAM,YAAY,CAAC,kDAAkD,CAAC,CAAC;IACpG,IAAI,OAAO,CAAC,cAAc,IAAI,OAAO,CAAC,UAAU,IAAI,OAAO,CAAC,kBAAkB,IAAI,OAAO,CAAC,iBAAiB;QAAE,MAAM,YAAY,CAAC,uEAAuE,CAAC,CAAC;IACzM,MAAM,MAAM,GAAG,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC,MAAM,aAAa,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,MAAM,iBAAiB,CAAC,OAAO,CAAC,aAAc,CAAC,CAAC;IAC5H,wBAAwB,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,qBAAqB,CAAC,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC,CAAC,UAAU,EACnI,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;IAC/E,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,IAAI,SAAS,EAAE,CAAC;IAC7C,MAAM,QAAQ,GAAG,MAAM,gBAAgB,CAAC,MAAM,EAAE,OAAO,CAAC,IAAI,CAAC,CAAC;IAC9D,MAAM,OAAO,GAAG;QACd,GAAG,OAAO,CAAC,OAAO,EAAE,OAAO,EAAE,QAAQ,CAAC,KAAK;QAC3C,QAAQ,EAAE,OAAO,CAAC,OAAO,CAAC,QAAQ,IAAI,MAAM,CAAC,OAAO,CAAC,QAAQ,CAAC,iBAAiB;QAC/E,cAAc,EAAE,OAAO,CAAC,OAAO,CAAC,cAAc,IAAI,CAAC;QACnD,8EAA8E;QAC9E,UAAU,EAAE,OAAO,CAAC,OAAO,CAAC,UAAU,IAAI,CAAC;QAC3C,gBAAgB,EAAE,OAAO,CAAC,OAAO,CAAC,gBAAgB,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,gBAAgB;QAC5F,sBAAsB,EAAE,OAAO,CAAC,OAAO,CAAC,sBAAsB,IAAI,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,sBAAsB;KAChH,CAAC;IACF,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,MAAM,EAAE,WAAW,CAAC,CAAC;IACjF,MAAM,iBAAiB,GAAG,MAAM,mBAAmB,CAAC,OAAO,CAAC,CAAC;IAC7D,IAAI,iBAAiB,CAAC,YAAY,KAAK,MAAM,CAAC,QAAQ,CAAC,EAAE;QAAE,MAAM,YAAY,CAAC,+CAA+C,CAAC,CAAC;IAC/H,OAAO,OAAO,CAAC,EAAE,GAAG,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,iBAAiB,EAAE,iBAAiB,EAAE,qBAAqB,EAAE,cAAc,EAAE,KAAK,EAAE,KAAK,EAAE,IAAI,EAAE,EAAE;YAC/I,IAAI,KAAK,KAAK,UAAU,EAAE,CAAC;gBACzB,MAAM,SAAS,CAAC,SAAS,CAAC,CAAC;gBAC3B,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,qBAAqB,CAAC,EAAE,MAAM,CAAC,IAAI,CAAC,CAAC;gBAChF,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,eAAe,CAAC,EAAE,MAAM,CAAC,QAAQ,CAAC,CAAC;gBAC9E,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,wBAAwB,CAAC,EAAE;oBACpE,GAAG,MAAM,CAAC,OAAO;oBACjB,SAAS,EAAE,EAAE,QAAQ,EAAE,OAAO,CAAC,QAAQ,EAAE,cAAc,EAAE,OAAO,CAAC,cAAc,EAAE,UAAU,EAAE,OAAO,CAAC,UAAU,EAAE,qBAAqB,EAAE,KAAK,EAAE;iBAChJ,CAAC,CAAC;gBACH,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,cAAc,CAAC,EAAE,EAAE,MAAM,EAAE,QAAQ,CAAC,MAAM,EAAE,KAAK,EAAE,QAAQ,CAAC,KAAK,EAAE,cAAc,EAAE,MAAM,CAAC,IAAI,CAAC,cAAc,EAAE,eAAe,EAAE,QAAQ,CAAC,MAAM,EAAE,CAAC,CAAC;YAChM,CAAC;YACD,MAAM,OAAO,CAAC,cAAc,EAAE,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC;QAC9C,CAAC,EAAE,CAAC,CAAC;AACP,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,gBAAgB,CAAC,MAAyB,EAAE,IAAY;IAC5E,MAAM,MAAM,GAAG,UAAU,CAAC,EAAE,IAAI,EAAE,MAAM,CAAC,IAAI,EAAE,QAAQ,EAAE,2BAA2B,EAAE,CAAC,CAAC;IACxF,MAAM,KAAK,GAAG,MAAM,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,OAAO,EAAE,YAAY,CAAC,CAAC,CAAC;IACtE,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;IACjD,MAAM,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,EAAE,OAAO,CAAC,EAAE,MAAM,EAAE,KAAK,IAAI,EAAE;QAC/D,IAAI,CAAC;YACH,MAAM,QAAQ,GAAG,MAAM,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE,QAAQ,CAAC,CAAC,CAAC;YAClE,IAAI,UAAU,CAAC,QAAQ,CAAC,IAAI,CAAC,KAAK,UAAU,CAAC,MAAM,CAAC,IAAI,CAAC;gBAAE,MAAM,YAAY,CAAC,qCAAqC,CAAC,CAAC;YACrH,MAAM,QAAQ,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE,eAAe,CAAC,EAAE,MAAM,CAAC,CAA6C,CAAC;YACpI,IAAI,QAAQ,CAAC,MAAM,KAAK,MAAM;gBAAE,MAAM,YAAY,CAAC,6CAA6C,CAAC,CAAC;YAClG,IAAI,QAAQ,CAAC,YAAY,KAAK,MAAM,mBAAmB,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;gBAAE,MAAM,YAAY,CAAC,wCAAwC,CAAC,CAAC;YAClJ,OAAO;QACT,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAAC,IAAK,KAA+B,CAAC,IAAI,KAAK,QAAQ;gBAAE,MAAM,KAAK,CAAC;QAAC,CAAC;QACxF,MAAM,SAAS,GAAG,MAAM,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,EAAE,YAAY,CAAC,CAAC,CAAC;QAChE,IAAI,CAAC;YACH,MAAM,EAAE,CAAC,MAAM,CAAC,SAAS,EAAE,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,QAAQ,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,WAAW,EAAE,KAAK,EAAE,CAAC,CAAC;YACpG,MAAM,QAAQ,GAAG,MAAM,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,QAAQ,CAAC,CAAC,CAAC;YACrE,IAAI,UAAU,CAAC,QAAQ,CAAC,IAAI,CAAC,KAAK,UAAU,CAAC,MAAM,CAAC,IAAI,CAAC;gBAAE,MAAM,YAAY,CAAC,sCAAsC,CAAC,CAAC;YACtH,MAAM,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,OAAO,CAAC,CAAC,CAAC;YAC3C,KAAK,MAAM,IAAI,IAAI,QAAQ,CAAC,KAAK,EAAE,CAAC;gBAClC,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,CAAC,CAAC;gBACxD,MAAM,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,SAAS,EAAE,IAAI,CAAC,IAAI,CAAC,EAAE,QAAQ,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;gBAClF,MAAM,YAAY,GAAG,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,KAAK,aAAa,IAAI,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,aAAa,CAAC;gBAC1G,IAAI,qBAAqB,GAAG,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,qBAAqB,GAAG,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,gBAAgB,CAAC;gBACjH,IAAI,YAAY,EAAE,CAAC;oBACjB,oEAAoE;oBACpE,qEAAqE;oBACrE,+DAA+D;oBAC/D,+DAA+D;oBAC/D,qEAAqE;oBACrE,MAAM,WAAW,GAAG,YAAY,CAAC,QAAQ,KAAK,8BAA8B,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;oBACrF,qBAAqB,GAAG,WAAW,GAAG,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,qBAAqB,GAAG,CAAC,GAAG,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,gBAAgB,CAAC;gBACjI,CAAC;gBACD,kEAAkE;gBAClE,sEAAsE;gBACtE,MAAM,SAAS,GAAG,eAAe,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;gBAC/C,MAAM,KAAK,GAAG,SAAS,CAAC,KAAgC,CAAC;gBACzD,KAAK,CAAC,WAAW,GAAG,MAAM,CAAC,KAAK,CAAC,WAAW,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,qBAAqB,GAAG,IAAI,CAAC,CAAC;gBACxF,MAAM,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,WAAW,CAAC,EAAE,aAAa,CAAC,SAAS,CAAC,CAAC,CAAC;gBAC5E,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,uBAAuB,CAAC,EAAE;oBAClE,cAAc,EAAE,GAAG,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,EAAE,IAAI,EAAE,IAAI,CAAC,MAAM;oBACxD,cAAc,EAAE,MAAM,CAAC,IAAI,CAAC,cAAc,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO;oBACnE,cAAc,EAAE,MAAM,CAAC,QAAQ,CAAC,cAAc,EAAE,OAAO,EAAE,MAAM,CAAC,QAAQ,CAAC,OAAO;oBAChF,cAAc,EAAE,EAAE,WAAW,EAAE,MAAM,CAAC,QAAQ,CAAC,cAAc,EAAE;oBAC/D,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE,iBAAiB,EAAG,IAAI,CAAC,MAAM,CAAC,KAAiC,CAAC,WAAW;oBAChG,6BAA6B,EAAE,qBAAqB;oBACpD,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,KAAK,YAAY,CAAC,CAAC,CAAC,EAAE,eAAe,EAAE,IAAI,CAAC,KAAK,CAAC,MAAM,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,SAAS,EAAE,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,MAAM,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;oBAC5K,cAAc,EAAE,MAAM,CAAC,IAAI,CAAC,cAAc,EAAE,WAAW,EAAE,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,KAAK,IAAI,CAAC,EAAE,CAAE,CAAC,WAAW;iBAC3H,CAAC,CAAC;YACL,CAAC;YACD,MAAM,OAAO,GAAG,MAAM,CAAC,QAAQ,CAAC,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,cAAc,CAAE,CAAC;YACzE,MAAM,gBAAgB,GAAG,MAAM,6BAA6B,CAAC;gBAC3D,OAAO,EAAE,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,OAAO,CAAC;gBACtC,SAAS,EAAE,EAAE,EAAE,EAAE,MAAM,CAAC,QAAQ,CAAC,EAAE,EAAE,QAAQ,EAAE,MAAM,CAAC,IAAI,CAAC,cAAc,EAAE;gBAC3E,OAAO,EAAE,EAAE,EAAE,EAAE,2BAA2B,EAAE,QAAQ,EAAE,MAAM,EAAE,eAAe,EAAE,4BAA4B,EAAE;gBAC7G,OAAO,EAAE;oBACP,WAAW,EAAE;wBACX,aAAa,EAAE,MAAM,CAAC,QAAQ,CAAC,cAAc;wBAC7C,SAAS,EAAE,OAAO,CAAC,SAAS;wBAC5B,KAAK,EAAE,OAAO,CAAC,KAAK;wBACpB,OAAO,EAAE,iBAAiB;qBAC3B;iBACF;gBACD,OAAO,EAAE,QAAQ,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC;aAC/C,CAAC,CAAC;YACH,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,OAAO,EAAE,wBAAwB,CAAC,EAAE,gBAAgB,CAAC,CAAC;YACjG,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,QAAQ,EAAE,qBAAqB,CAAC,EAAE,MAAM,CAAC,IAAI,CAAC,CAAC;YAC1F,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,eAAe,CAAC,EAAE,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,mBAAmB,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,OAAO,CAAC,CAAC,EAAE,CAAC,CAAC;YACjJ,MAAM,MAAM,CAAC,SAAS,EAAE,MAAM,CAAC,CAAC;QAClC,CAAC;gBAAS,CAAC;YAAC,MAAM,EAAE,CAAC,SAAS,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;QAAC,CAAC;IACtE,CAAC,CAAC,CAAC;IACH,OAAO,EAAE,MAAM,EAAE,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE,QAAQ,CAAC,EAAE,KAAK,EAAE,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE,OAAO,CAAC,EAAE,MAAM,EAAE,CAAC;AAC5F,CAAC;AAED,mFAAmF;AACnF,MAAM,CAAC,KAAK,UAAU,8BAA8B,CAAC,gBAAwB,EAAE,eAAuB;IACpG,MAAM,MAAM,GAAG,IAAI,CAAC,OAAO,CAAC,eAAe,CAAC,CAAC;IAC7C,MAAM,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACvD,MAAM,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,yCAAyC;IAC9D,MAAM,aAAa,GAAG,MAAM,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,0BAA0B,CAAC,CAAC,CAAC;IACjG,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,MAAM,aAAa,CAAC,gBAAgB,CAAC,CAAC;QACrD,MAAM,QAAQ,GAAG,MAAM,gBAAgB,CAAC,MAAM,EAAE,aAAa,CAAC,CAAC;QAC/D,KAAK,MAAM,KAAK,IAAI,MAAM,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC;YAClD,MAAM,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,KAAK,EAAE,KAAK,CAAC,EAAE,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE,KAAK,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,KAAK,EAAE,KAAK,EAAE,CAAC,CAAC;QAC9H,CAAC;QACD,MAAM,QAAQ,GAAG,MAAM,4BAA4B,CAAC,MAAM,CAAC,CAAC;QAC5D,IAAI,CAAC,QAAQ;YAAE,MAAM,YAAY,CAAC,wDAAwD,CAAC,CAAC;QAC5F,OAAO;YACL,OAAO,EAAE,MAAM;YACf,YAAY,EAAE,QAAQ,CAAC,SAAS,CAAC,EAAE;YACnC,kBAAkB,EAAE,QAAQ,CAAC,cAAc;YAC3C,qBAAqB,EAAE,MAAM,CAAC,IAAI,CAAC,cAAc;YACjD,KAAK,EAAE,QAAQ,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,OAAO,CAAC;YACjD,OAAO,EAAE,QAAQ,CAAC,OAAO;SAC1B,CAAC;IACJ,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,MAAM,EAAE,CAAC,MAAM,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;QACnD,MAAM,KAAK,CAAC;IACd,CAAC;YAAS,CAAC;QACT,MAAM,EAAE,CAAC,aAAa,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;IAC5D,CAAC;AACH,CAAC"}
|
|
@@ -116,6 +116,7 @@ async function collectSlot(input, executionPlan, resolution, runtimeId, slot, ex
|
|
|
116
116
|
publicationMode: "replace-invalid",
|
|
117
117
|
runtimeId,
|
|
118
118
|
env: input.env ?? process.env,
|
|
119
|
+
...(input.signal ? { signal: input.signal } : {}),
|
|
119
120
|
requireCompleteMarker: true,
|
|
120
121
|
}, trial, index, imported);
|
|
121
122
|
if (!imported.some((entry) => entry.trial_id === ref.trial_id))
|
|
@@ -138,7 +139,7 @@ async function collectSlot(input, executionPlan, resolution, runtimeId, slot, ex
|
|
|
138
139
|
ref,
|
|
139
140
|
source: {
|
|
140
141
|
source_trial_id: ref.trial_id,
|
|
141
|
-
source_run_id: ref.run_id,
|
|
142
|
+
...(ref.run_group ? { source_run_group_id: ref.run_group.run_group_id } : { source_run_id: ref.run_id }),
|
|
142
143
|
source_work_id: work.work_id,
|
|
143
144
|
source_backend_directory: relativeBackend.split(path.sep).join("/"),
|
|
144
145
|
},
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"collect-only-rerun.js","sourceRoot":"","sources":["../../../src/evals/collect-only-rerun.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,kBAAkB,CAAC;AAC3C,OAAO,IAAI,MAAM,WAAW,CAAC;AAE7B,OAAO,EAAE,mBAAmB,EAAE,MAAM,sBAAsB,CAAC;AAE3D,OAAO,EAAE,UAAU,EAAE,cAAc,EAAE,eAAe,EAAE,QAAQ,EAAE,MAAM,wBAAwB,CAAC;AAC/F,OAAO,EAAE,kBAAkB,EAAE,MAAM,qBAAqB,CAAC;AACzD,OAAO,EAAE,sBAAsB,EAAE,MAAM,qBAAqB,CAAC;AAC7D,OAAO,EAAE,+BAA+B,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AACnF,OAAO,EAAE,kBAAkB,EAAE,MAAM,kBAAkB,CAAC;AAEtD,OAAO,EAAE,UAAU,EAAE,iBAAiB,EAAE,OAAO,EAAE,SAAS,EAAE,WAAW,EAAE,MAAM,kBAAkB,CAAC;AAElG,OAAO,EAAE,kBAAkB,EAAE,uBAAuB,EAAE,2BAA2B,EAAE,MAAM,mBAAmB,CAAC;
|
|
1
|
+
{"version":3,"file":"collect-only-rerun.js","sourceRoot":"","sources":["../../../src/evals/collect-only-rerun.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,kBAAkB,CAAC;AAC3C,OAAO,IAAI,MAAM,WAAW,CAAC;AAE7B,OAAO,EAAE,mBAAmB,EAAE,MAAM,sBAAsB,CAAC;AAE3D,OAAO,EAAE,UAAU,EAAE,cAAc,EAAE,eAAe,EAAE,QAAQ,EAAE,MAAM,wBAAwB,CAAC;AAC/F,OAAO,EAAE,kBAAkB,EAAE,MAAM,qBAAqB,CAAC;AACzD,OAAO,EAAE,sBAAsB,EAAE,MAAM,qBAAqB,CAAC;AAC7D,OAAO,EAAE,+BAA+B,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AACnF,OAAO,EAAE,kBAAkB,EAAE,MAAM,kBAAkB,CAAC;AAEtD,OAAO,EAAE,UAAU,EAAE,iBAAiB,EAAE,OAAO,EAAE,SAAS,EAAE,WAAW,EAAE,MAAM,kBAAkB,CAAC;AAElG,OAAO,EAAE,kBAAkB,EAAE,uBAAuB,EAAE,2BAA2B,EAAE,MAAM,mBAAmB,CAAC;AA+B7G,MAAM,CAAC,KAAK,UAAU,oBAAoB,CAAC,KAA4B;IACrE,MAAM,aAAa,GAAG,sBAAsB,CAAC,MAAM,QAAQ,CAAU,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,qBAAqB,CAAC,CAAC,CAAC,CAAC;IAC7H,IAAI,aAAa,CAAC,OAAO,KAAK,KAAK,CAAC,MAAM,IAAI,aAAa,CAAC,UAAU,KAAK,OAAO;QAAE,MAAM,WAAW,CAAC,mDAAmD,CAAC,CAAC;IAC3J,MAAM,UAAU,GAAG,MAAM,QAAQ,CAAmB,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,iBAAiB,CAAC,CAAC,CAAC;IACvG,IAAI,UAAU,CAAC,QAAQ,KAAK,cAAc,CAAC,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,iBAAiB,EAAE,6BAA6B,CAAC;QAAE,MAAM,WAAW,CAAC,0CAA0C,CAAC,CAAC;IACjL,MAAM,SAAS,GAAG,cAAc,CAAC,KAAK,CAAC,IAAI,CAAC,iBAAiB,CAAC,UAAU,EAAE,uBAAuB,CAAC,CAAC;IACnG,IAAI,QAAQ,GAAG,KAAK,CAAC,QAAQ,CAAC;IAC9B,MAAM,QAAQ,GAAG,QAAQ,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,cAAc,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,KAAK,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;IACvH,MAAM,aAAa,GAAqB,EAAE,CAAC;IAC3C,MAAM,OAAO,GAAsB,EAAE,CAAC;IACtC,KAAK,MAAM,IAAI,IAAI,KAAK,CAAC,cAAc,EAAE,CAAC;QACxC,MAAM,SAAS,GAAG,MAAM,WAAW,CAAC,KAAK,EAAE,aAAa,EAAE,UAAU,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,GAAG,QAAQ,EAAE,GAAG,aAAa,CAAC,CAAC,CAAC;QACxH,aAAa,CAAC,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC;QAClC,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;IACjC,CAAC;IACD,KAAK,MAAM,GAAG,IAAI,aAAa;QAAE,QAAQ,GAAG,+BAA+B,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;IAC3F,IAAI,QAAQ,CAAC,UAAU,KAAK,KAAK,CAAC,QAAQ,CAAC,UAAU;QAAE,MAAM,iBAAiB,CAAC,KAAK,CAAC,aAAa,EAAE,QAAQ,CAAC,CAAC;IAC9G,MAAM,eAAe,GAAG,iBAAiB,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,QAAQ,EAAE,QAAQ,CAAC,CAAC;IAC3F,MAAM,cAAc,GAAG,WAAW,CAAC,eAAe,CAAC,CAAC;IACpD,MAAM,cAAc,GAAG,SAAS,CAAC,aAAa,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC,EAAE,OAAO,EAAE,GAAG,CAAC,OAAO,EAAE,OAAO,EAAE,GAAG,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,CAAC;IAC/G,MAAM,aAAa,GAAG,WAAW,CAAC,cAAc,CAAC,CAAC;IAClD,MAAM,SAAS,GAAG,eAAe,CAAC,MAAM,KAAK,CAAC,CAAC;IAC/C,MAAM,WAAW,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC7C,MAAM,eAAe,CAAC,KAAK,EAAE,QAAQ,EAAE,eAAe,EAAE,SAAS,EAAE,WAAW,CAAC,CAAC;IAChF,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,cAAc,EAAE,YAAY,CAAC,EAAE;QACnE,cAAc,EAAE,cAAc;QAC9B,QAAQ,EAAE,KAAK,CAAC,OAAO;QACvB,OAAO,EAAE,KAAK,CAAC,MAAM;QACrB,UAAU,EAAE,cAAc;QAC1B,SAAS,EAAE,kBAAkB,CAAC,cAAc,CAAC;QAC7C,MAAM,EAAE,WAAW;QACnB,KAAK,EAAE,WAAW,CAAC,KAAK,CAAC,cAAc,CAAC;QACxC,MAAM,EAAE,SAAS,CAAC,KAAK,CAAC,cAAc,CAAC;QACvC,cAAc,EAAE,aAAa;QAC7B,eAAe,EAAE,cAAc;QAC/B,uBAAuB,EAAE,cAAc;QACvC,wBAAwB,EAAE,eAAe;QACzC,OAAO;QACP,WAAW,EAAE,SAAS,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,QAAQ;QAC/C,UAAU,EAAE,KAAK,CAAC,SAAS;QAC3B,YAAY,EAAE,WAAW;QACzB,UAAU,EAAE,WAAW;KACxB,CAAC,CAAC;IACH,OAAO;QACL,cAAc,EAAE,GAAG;QACnB,IAAI,EAAE,YAAY;QAClB,QAAQ,EAAE,KAAK,CAAC,OAAO;QACvB,UAAU,EAAE,cAAc;QAC1B,SAAS,EAAE,kBAAkB,CAAC,cAAc,CAAC;QAC7C,OAAO,EAAE,KAAK,CAAC,MAAM;QACrB,MAAM,EAAE,WAAW;QACnB,cAAc,EAAE,WAAW,CAAC,KAAK,CAAC,cAAc,CAAC;QACjD,cAAc,EAAE,aAAa;QAC7B,uBAAuB,EAAE,cAAc;QACvC,eAAe,EAAE,SAAS,CAAC,KAAK,CAAC,cAAc,CAAC;QAChD,eAAe,EAAE,cAAc;QAC/B,wBAAwB,EAAE,eAAe;QACzC,OAAO;QACP,WAAW,EAAE,SAAS,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,QAAQ;QAC/C,UAAU,EAAE,KAAK,CAAC,SAAS;QAC3B,YAAY,EAAE,WAAW;KAC1B,CAAC;AACJ,CAAC;AAED,KAAK,UAAU,WAAW,CACxB,KAA4B,EAC5B,aAAkC,EAClC,UAA4B,EAC5B,SAAiB,EACjB,IAAmB,EACnB,YAA8B;IAE9B,MAAM,WAAW,GAAG,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,OAAO,KAAK,IAAI,CAAC,OAAO,IAAI,KAAK,CAAC,OAAO,KAAK,IAAI,CAAC,OAAO,CAAC,CAAC;IAC1H,MAAM,IAAI,GAAG,WAAW,IAAI,aAAa,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,KAAK,CAAC,QAAQ,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC;IAChH,IAAI,CAAC,WAAW,IAAI,CAAC,IAAI,IAAI,IAAI,CAAC,QAAQ,CAAC,MAAM,KAAK,CAAC,IAAI,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,KAAK,IAAI,CAAC,OAAO,EAAE,CAAC;QAC7F,MAAM,WAAW,CAAC,8CAA8C,UAAU,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACtF,CAAC;IACD,MAAM,aAAa,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,QAAQ,EAAE,YAAY,EAAE,IAAI,CAAC,OAAO,CAAC,CAAC;IAC3F,IAAI,OAAO,CAAC;IACZ,IAAI,CAAC;QAAC,OAAO,GAAG,MAAM,OAAO,CAAC,aAAa,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC;IAAC,CAAC;IACxE,OAAO,KAAK,EAAE,CAAC;QACb,IAAK,KAA+B,CAAC,IAAI,KAAK,QAAQ;YAAE,MAAM,WAAW,CAAC,sCAAsC,UAAU,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACpI,MAAM,KAAK,CAAC;IACd,CAAC;IACD,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,WAAW,EAAE,IAAI,kBAAkB,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;SACjG,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,CAAC,OAAO,EAAE,CAAC;IAC/C,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;QAC3B,MAAM,eAAe,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,YAAY,EAAE,IAAI,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC;QAC/E,MAAM,YAAY,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,eAAe,EAAE,KAAK,CAAC,CAAC;QAC5E,MAAM,GAAG,GAAG,MAAM,mBAAmB,CAAC,YAAY,CAAC,CAAC;QACpD,MAAM,MAAM,GAAG,KAAK,CAAC,OAAO,CAAC,GAAG,EAAE,aAAa,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,aAA0C,CAAC,CAAC,CAAC,EAAE,CAAC;QACvG,IAAI,CAAC,GAAG,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;YAAE,SAAS;QAC1C,MAAM,QAAQ,GAAG,CAAC,GAAG,YAAY,CAAC,CAAC;QACnC,KAAK,MAAM,CAAC,KAAK,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,EAAE,EAAE,CAAC;YAC9C,IAAI,CAAC;gBACH,MAAM,GAAG,GAAG,MAAM,kBAAkB,CAAC;oBACnC,IAAI,EAAE,KAAK,CAAC,IAAI;oBAChB,MAAM,EAAE,KAAK,CAAC,MAAM;oBACpB,aAAa,EAAE,KAAK,CAAC,aAAa;oBAClC,kBAAkB,EAAE,YAAY;oBAChC,eAAe,EAAE,IAAI,CAAC,OAAO;oBAC7B,OAAO,EAAE,KAAK,CAAC,OAAO;oBACtB,gBAAgB,EAAE,UAAU;oBAC5B,WAAW,EAAE,KAAK,CAAC,OAAO,CAAC,YAAY;oBACvC,iBAAiB,EAAE,KAAK,CAAC,OAAO,CAAC,kBAAkB;oBACnD,eAAe,EAAE,iBAAiB;oBAClC,SAAS;oBACT,GAAG,EAAE,KAAK,CAAC,GAAG,IAAI,OAAO,CAAC,GAAG;oBAC7B,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;oBACjD,qBAAqB,EAAE,IAAI;iBAC5B,EAAE,KAAK,EAAE,KAAK,EAAE,QAAQ,CAAC,CAAC;gBAC3B,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,QAAQ,KAAK,GAAG,CAAC,QAAQ,CAAC;oBAAE,QAAQ,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;YACrF,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,IAAI,KAAK,YAAY,uBAAuB;oBAAE,SAAS;gBACvD,MAAM,KAAK,CAAC;YACd,CAAC;QACH,CAAC;QACD,MAAM,GAAG,GAAG,QAAQ,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,OAAO,KAAK,IAAI,CAAC,OAAO,IAAI,KAAK,CAAC,OAAO,KAAK,IAAI,CAAC,OAAO,IAAI,KAAK,CAAC,kBAAkB,KAAK,OAAO,CAAC,CAAC;QAC/I,IAAI,CAAC,GAAG;YAAE,SAAS;QACnB,MAAM,2BAA2B,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,EAAE;YACjE,WAAW,EAAE,KAAK,CAAC,OAAO,CAAC,YAAY;YACvC,iBAAiB,EAAE,KAAK,CAAC,OAAO,CAAC,kBAAkB;SACpD,CAAC,CAAC;QACH,OAAO;YACL,GAAG;YACH,MAAM,EAAE;gBACN,eAAe,EAAE,GAAG,CAAC,QAAQ;gBAC7B,GAAG,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,mBAAmB,EAAE,GAAG,CAAC,SAAS,CAAC,YAAY,EAAE,CAAC,CAAC,CAAC,EAAE,aAAa,EAAE,GAAG,CAAC,MAAM,EAAE,CAAC;gBACxG,cAAc,EAAE,IAAI,CAAC,OAAO;gBAC5B,wBAAwB,EAAE,eAAe,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC;aACpE;SACF,CAAC;IACJ,CAAC;IACD,MAAM,IAAI,UAAU,CAAC,mDAAmD,UAAU,CAAC,IAAI,CAAC,EAAE,EAAE;QAC1F,IAAI,EAAE,sCAAsC;QAC5C,QAAQ,EAAE,EAAE;KACb,CAAC,CAAC;AACL,CAAC;AAED,KAAK,UAAU,eAAe,CAC5B,KAA4B,EAC5B,QAAwB,EACxB,eAAgC,EAChC,SAAkB,EAClB,WAAmB;IAEnB,MAAM,MAAM,GAA4B;QACtC,GAAG,CAAC,KAAK,CAAC,cAAc,IAAI,EAAE,CAAC;QAC/B,cAAc,EAAE,cAAc;QAC9B,OAAO,EAAE,KAAK,CAAC,MAAM;QACrB,MAAM,EAAE,SAAS,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,QAAQ;QAC1C,SAAS,EAAE,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE;QAC7B,SAAS,EAAE,KAAK,CAAC,IAAI,CAAC,SAAS;QAC/B,OAAO,EAAE,KAAK,CAAC,OAAO,CAAC,OAAO;QAC9B,YAAY,EAAE,KAAK,CAAC,OAAO,CAAC,YAAY;QACxC,kBAAkB,EAAE,KAAK,CAAC,OAAO,CAAC,kBAAkB;QACpD,UAAU,EAAE,QAAQ,CAAC,UAAU;QAC/B,MAAM,EAAE,QAAQ,CAAC,MAAM;QACvB,OAAO,EAAE,kBAAkB,CAAC,QAAQ,CAAC,MAAM,CAAC;QAC5C,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,wBAAwB,EAAE,OAAO,EAAE,uCAAuC,eAAe,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,EAAE,EAAE,CAAC;QACjK,UAAU,EAAE,OAAO,KAAK,CAAC,cAAc,EAAE,UAAU,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,cAAc,CAAC,UAAU,CAAC,CAAC,CAAC,QAAQ,CAAC,UAAU;QACxH,YAAY,EAAE,WAAW;KAC1B,CAAC;IACF,IAAI,SAAS;QAAE,OAAO,MAAM,CAAC,KAAK,CAAC;IACnC,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,aAAa,CAAC,EAAE,MAAM,CAAC,CAAC;AAC/E,CAAC;AAED,SAAS,cAAc,CAAC,KAAc,EAAE,KAAa;IACnD,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,CAAC,KAAK;QAAE,MAAM,WAAW,CAAC,GAAG,KAAK,aAAa,CAAC,CAAC;IAClF,OAAO,KAAK,CAAC;AACf,CAAC;AAED,SAAS,WAAW,CAAC,OAAe;IAClC,OAAO,IAAI,UAAU,CAAC,OAAO,EAAE,EAAE,IAAI,EAAE,+BAA+B,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC,CAAC;AACzF,CAAC"}
|
|
@@ -10,7 +10,7 @@ export async function prepareEvalDirectory(input) {
|
|
|
10
10
|
if (JSON.stringify(existingRequest) !== JSON.stringify(input.request)) {
|
|
11
11
|
throw new HitchError(`precreated eval request does not match: ${input.evalId}`, { code: "eval_request_conflict", exitCode: 2 });
|
|
12
12
|
}
|
|
13
|
-
if (await readJSON(path.join(directory, "result.json"), null)) {
|
|
13
|
+
if (!input.replaceTerminal && await readJSON(path.join(directory, "result.json"), null)) {
|
|
14
14
|
throw new HitchError(`eval is already terminal: ${input.evalId}`, { code: "eval_id_conflict", exitCode: 2 });
|
|
15
15
|
}
|
|
16
16
|
return directory;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"directory.js","sourceRoot":"","sources":["../../../src/evals/directory.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,EAAE,MAAM,kBAAkB,CAAC;AACzC,OAAO,IAAI,MAAM,WAAW,CAAC;AAE7B,OAAO,EAAE,UAAU,EAAE,eAAe,EAAE,QAAQ,EAAE,MAAM,wBAAwB,CAAC;AAE/E,MAAM,CAAC,KAAK,UAAU,oBAAoB,CAAC,
|
|
1
|
+
{"version":3,"file":"directory.js","sourceRoot":"","sources":["../../../src/evals/directory.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,EAAE,MAAM,kBAAkB,CAAC;AACzC,OAAO,IAAI,MAAM,WAAW,CAAC;AAE7B,OAAO,EAAE,UAAU,EAAE,eAAe,EAAE,QAAQ,EAAE,MAAM,wBAAwB,CAAC;AAE/E,MAAM,CAAC,KAAK,UAAU,oBAAoB,CAAC,KAM1C;IACC,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,cAAc,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;IAChE,IAAI,KAAK,CAAC,UAAU,EAAE,CAAC;QACrB,MAAM,eAAe,GAAG,MAAM,QAAQ,CAAqB,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,cAAc,CAAC,EAAE,IAAI,CAAC,CAAC;QACvG,IAAI,CAAC,eAAe;YAAE,MAAM,IAAI,UAAU,CAAC,uCAAuC,KAAK,CAAC,MAAM,EAAE,EAAE,EAAE,IAAI,EAAE,gBAAgB,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC,CAAC;QAC3I,IAAI,IAAI,CAAC,SAAS,CAAC,eAAe,CAAC,KAAK,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,OAAO,CAAC,EAAE,CAAC;YACtE,MAAM,IAAI,UAAU,CAAC,2CAA2C,KAAK,CAAC,MAAM,EAAE,EAAE,EAAE,IAAI,EAAE,uBAAuB,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC,CAAC;QAClI,CAAC;QACD,IAAI,CAAC,KAAK,CAAC,eAAe,IAAI,MAAM,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,aAAa,CAAC,EAAE,IAAI,CAAC,EAAE,CAAC;YACxF,MAAM,IAAI,UAAU,CAAC,6BAA6B,KAAK,CAAC,MAAM,EAAE,EAAE,EAAE,IAAI,EAAE,kBAAkB,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC,CAAC;QAC/G,CAAC;QACD,OAAO,SAAS,CAAC;IACnB,CAAC;IACD,IAAI,CAAC;QACH,MAAM,KAAK,CAAC,SAAS,EAAE,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,CAAC;IAC1C,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAK,KAA+B,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;YACvD,MAAM,IAAI,UAAU,CAAC,2BAA2B,KAAK,CAAC,MAAM,EAAE,EAAE,EAAE,IAAI,EAAE,kBAAkB,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC,CAAC;QAC7G,CAAC;QACD,MAAM,KAAK,CAAC;IACd,CAAC;IACD,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,cAAc,CAAC,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC;IAC3E,OAAO,SAAS,CAAC;AACnB,CAAC"}
|