@tangle-network/agent-eval 0.122.2 → 0.122.4
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/dist/analyst/index.js +3 -4
- package/dist/analyst/index.js.map +1 -1
- package/dist/benchmarks/index.js +5 -6
- package/dist/campaign/index.d.ts +273 -252
- package/dist/campaign/index.js +8 -5
- package/dist/{chunk-N3QPYIVG.js → chunk-4R4GTZBZ.js} +10 -10
- package/dist/{chunk-EXMKNOP5.js → chunk-75OCWDXJ.js} +137 -43
- package/dist/chunk-75OCWDXJ.js.map +1 -0
- package/dist/{chunk-BD2OK6ZW.js → chunk-7GLJX32M.js} +2 -2
- package/dist/{chunk-5CVUPHJ4.js → chunk-EAWKAVID.js} +74 -1
- package/dist/chunk-EAWKAVID.js.map +1 -0
- package/dist/{chunk-OZFTLMGS.js → chunk-GTSVLG2M.js} +2 -2
- package/dist/{chunk-U4R2AGY5.js → chunk-MWN2X7C5.js} +42 -32
- package/dist/{chunk-U4R2AGY5.js.map → chunk-MWN2X7C5.js.map} +1 -1
- package/dist/{chunk-5BYTIDZ7.js → chunk-SFXDMFGV.js} +2 -2
- package/dist/{chunk-XKSDSGSC.js → chunk-ULW7AATT.js} +3 -5
- package/dist/{chunk-XKSDSGSC.js.map → chunk-ULW7AATT.js.map} +1 -1
- package/dist/contract/index.d.ts +3 -0
- package/dist/contract/index.js +5 -5
- package/dist/index.d.ts +36 -30
- package/dist/index.js +11 -11
- package/dist/index.js.map +1 -1
- package/dist/openapi.json +1 -1
- package/dist/{run-campaign-2YJ3Z6JS.js → run-campaign-J6RHX2VO.js} +2 -2
- package/package.json +1 -1
- package/dist/chunk-3YYRZDON.js +0 -45
- package/dist/chunk-3YYRZDON.js.map +0 -1
- package/dist/chunk-5CVUPHJ4.js.map +0 -1
- package/dist/chunk-EXMKNOP5.js.map +0 -1
- /package/dist/{chunk-N3QPYIVG.js.map → chunk-4R4GTZBZ.js.map} +0 -0
- /package/dist/{chunk-BD2OK6ZW.js.map → chunk-7GLJX32M.js.map} +0 -0
- /package/dist/{chunk-OZFTLMGS.js.map → chunk-GTSVLG2M.js.map} +0 -0
- /package/dist/{chunk-5BYTIDZ7.js.map → chunk-SFXDMFGV.js.map} +0 -0
- /package/dist/{run-campaign-2YJ3Z6JS.js.map → run-campaign-J6RHX2VO.js.map} +0 -0
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
import {
|
|
2
2
|
fsCampaignStorage,
|
|
3
3
|
runCampaign
|
|
4
|
-
} from "./chunk-
|
|
4
|
+
} from "./chunk-4R4GTZBZ.js";
|
|
5
5
|
import {
|
|
6
6
|
__export
|
|
7
7
|
} from "./chunk-PZ5AY32C.js";
|
|
@@ -763,4 +763,4 @@ export {
|
|
|
763
763
|
retrievalMetricsAtCutoff,
|
|
764
764
|
benchmarks_exports
|
|
765
765
|
};
|
|
766
|
-
//# sourceMappingURL=chunk-
|
|
766
|
+
//# sourceMappingURL=chunk-GTSVLG2M.js.map
|
|
@@ -10,11 +10,12 @@ import {
|
|
|
10
10
|
resolveRunDir,
|
|
11
11
|
runCampaign,
|
|
12
12
|
summarizeAgentReceiptIntegrity
|
|
13
|
-
} from "./chunk-
|
|
13
|
+
} from "./chunk-4R4GTZBZ.js";
|
|
14
14
|
import {
|
|
15
15
|
clamp01,
|
|
16
|
+
mapConcurrent,
|
|
16
17
|
validatePolicyEditCandidateRecord
|
|
17
|
-
} from "./chunk-
|
|
18
|
+
} from "./chunk-EAWKAVID.js";
|
|
18
19
|
import {
|
|
19
20
|
detectRewardHacking
|
|
20
21
|
} from "./chunk-ARU2PZFM.js";
|
|
@@ -2647,9 +2648,13 @@ function campaignBreakdown(campaign) {
|
|
|
2647
2648
|
// src/campaign/presets/run-optimization.ts
|
|
2648
2649
|
async function runOptimization(opts) {
|
|
2649
2650
|
const { proposer } = opts;
|
|
2651
|
+
const candidateConcurrency = opts.candidateConcurrency ?? 1;
|
|
2650
2652
|
if (typeof opts.runDir !== "string" || opts.runDir.trim().length === 0) {
|
|
2651
2653
|
throw new Error("runOptimization: runDir is required and must be a non-empty string");
|
|
2652
2654
|
}
|
|
2655
|
+
if (!Number.isInteger(candidateConcurrency) || candidateConcurrency < 1) {
|
|
2656
|
+
throw new Error("runOptimization: candidateConcurrency must be a positive integer");
|
|
2657
|
+
}
|
|
2653
2658
|
opts.runDir = resolveRunDir(opts.runDir, opts.repo);
|
|
2654
2659
|
const storage = opts.storage ?? fsCampaignStorage();
|
|
2655
2660
|
const costLedger = opts.costLedger ?? createRunCostLedger({
|
|
@@ -2749,34 +2754,39 @@ async function runOptimization(opts) {
|
|
|
2749
2754
|
const candidates = proposed.map(
|
|
2750
2755
|
(p) => isProposedCandidate(p) ? p : { surface: p, label: "", rationale: "" }
|
|
2751
2756
|
);
|
|
2752
|
-
const surfaceResults =
|
|
2753
|
-
|
|
2754
|
-
|
|
2755
|
-
|
|
2756
|
-
|
|
2757
|
-
|
|
2758
|
-
|
|
2759
|
-
|
|
2760
|
-
|
|
2761
|
-
|
|
2762
|
-
|
|
2763
|
-
|
|
2764
|
-
|
|
2765
|
-
|
|
2766
|
-
|
|
2767
|
-
|
|
2768
|
-
|
|
2769
|
-
|
|
2770
|
-
|
|
2771
|
-
|
|
2772
|
-
|
|
2773
|
-
|
|
2774
|
-
|
|
2775
|
-
|
|
2776
|
-
|
|
2777
|
-
|
|
2778
|
-
|
|
2779
|
-
|
|
2757
|
+
const surfaceResults = await mapConcurrent(
|
|
2758
|
+
candidates,
|
|
2759
|
+
candidateConcurrency,
|
|
2760
|
+
async ({ surface, label, rationale, candidateRecord }, i) => {
|
|
2761
|
+
const hash = surfaceHash(surface);
|
|
2762
|
+
const campaign = await runCampaign({
|
|
2763
|
+
...opts,
|
|
2764
|
+
costLedger,
|
|
2765
|
+
costPhase: "search.candidate",
|
|
2766
|
+
dispatch: (scenario, ctx) => opts.dispatchWithSurface(surface, scenario, ctx),
|
|
2767
|
+
runDir: `${opts.runDir}/gen-${gen}/candidate-${i}`
|
|
2768
|
+
});
|
|
2769
|
+
const composite = campaignMeanComposite(campaign);
|
|
2770
|
+
const coverage = campaignCoverage(
|
|
2771
|
+
campaign.cells,
|
|
2772
|
+
opts.scenarios,
|
|
2773
|
+
opts.reps ?? 1,
|
|
2774
|
+
requireJudgeScore
|
|
2775
|
+
);
|
|
2776
|
+
return {
|
|
2777
|
+
surfaceHash: hash,
|
|
2778
|
+
surface,
|
|
2779
|
+
label,
|
|
2780
|
+
rationale,
|
|
2781
|
+
...candidateRecord ? { candidateRecord } : {},
|
|
2782
|
+
campaign,
|
|
2783
|
+
composite,
|
|
2784
|
+
coverage
|
|
2785
|
+
};
|
|
2786
|
+
}
|
|
2787
|
+
);
|
|
2788
|
+
for (const result of surfaceResults) {
|
|
2789
|
+
const { surface, surfaceHash: hash, campaign, coverage, label, rationale } = result;
|
|
2780
2790
|
if (coverage.complete) {
|
|
2781
2791
|
scored.push(
|
|
2782
2792
|
toParetoParent(surface, hash, campaign, gen, label || void 0, rationale || void 0)
|
|
@@ -2988,7 +2998,7 @@ async function runImprovementLoop(opts) {
|
|
|
2988
2998
|
const dispatchTimeoutMs = opts.dispatchTimeoutMs ?? DEFAULT_DISPATCH_TIMEOUT_MS;
|
|
2989
2999
|
const optimization = await runOptimization({ ...opts, dispatchTimeoutMs, costLedger });
|
|
2990
3000
|
const winnerIsBaseline = optimization.winnerSurfaceHash === surfaceHash(opts.baselineSurface);
|
|
2991
|
-
const { runCampaign: runCampaign2 } = await import("./run-campaign-
|
|
3001
|
+
const { runCampaign: runCampaign2 } = await import("./run-campaign-J6RHX2VO.js");
|
|
2992
3002
|
const baselineOnHoldout = await runCampaign2({
|
|
2993
3003
|
...opts,
|
|
2994
3004
|
costLedger,
|
|
@@ -3780,4 +3790,4 @@ export {
|
|
|
3780
3790
|
provenanceSpansPath,
|
|
3781
3791
|
emitLoopProvenance
|
|
3782
3792
|
};
|
|
3783
|
-
//# sourceMappingURL=chunk-
|
|
3793
|
+
//# sourceMappingURL=chunk-MWN2X7C5.js.map
|