| { |
| "adapterParameterChecksum": "0c68b19cdbf8523c7bd57e2869e945a7a9df1511d3e07f711e95d33298af2c7e", |
| "baseModel": "google/gemma-4-E4B-it", |
| "baseRevision": "ee0ef6023621cff504d758262d4e04895a5af4a2", |
| "checkpointRunId": "alpino-e4b-v02-controller-escalation-32032090932", |
| "checkpointSubdir": "grpo-v5", |
| "freezePolicy": { |
| "frozenVersion": "v0.2", |
| "futureTrainingTrack": "v0.3", |
| "noFurtherWeightUpdates": true, |
| "note": "A.9 is the final evidence-bearing v0.2 checkpoint. Further GRPO must use a new version lineage and must not mutate or overwrite this checkpoint.", |
| "v03PrimaryFocus": "cross-entity reasoning" |
| }, |
| "heldoutDevelopment": { |
| "aooeSymbolicConformant": 6, |
| "artifactId": 9290659167, |
| "artifactZipSha256": "178795fe76ab2d74ab93b133ac52d5b94b50e55ff0e88c3b6896c052954f27de", |
| "attemptedPerArm": 6, |
| "benchmarkLoaded": false, |
| "immutableV2Loaded": false, |
| "promotionGatePassed": true, |
| "rawSymbolicConformant": 5, |
| "resultJsonSha256": "2680924f1cf5ed1d3a46608bfa0f175e95781cf4d42bceac2147e42aaeb16874", |
| "runId": 32035764957, |
| "sourceCommit": "57cdd537b45a6aa245497c5170ede0c8fbcef6ef", |
| "status": "passed", |
| "weightUpdates": false |
| }, |
| "immutableBenchmark": { |
| "artifactId": 9292050963, |
| "artifactZipSha256": "333edd6741ef33b54f4b680340e774721be8a1b8912450af8492e69527925b8a", |
| "branch": "feat/alpina-governed-content-track", |
| "evaluationPolicy": "single-successful-run-no-tuning-after-reveal", |
| "headSha": "cd497abd02683d00dfd0d06d2c5b42db9883129e", |
| "id": "alpina-immutable-v2", |
| "pairedDelta": { |
| "tokenReductionFraction": 0.7210104744300678, |
| "tokensPerValidTransitionReductionFraction": 0.7768083795440542, |
| "validTransitions": 4, |
| "validTransitionsPerPflopRatio": 4.480454946996467 |
| }, |
| "repository": "wordlift/aooe-benchmarks", |
| "resultJsonSha256": "6c3b089a2342fcf3f983998c9067a658f9a9a80ad893092cc2a852049afb9090", |
| "runId": 32041360569, |
| "scenarioCount": 24, |
| "seedRange": "72000-72023", |
| "status": "evaluated", |
| "summary": { |
| "aooe": { |
| "attempted": 24, |
| "correctAbstentions": 4, |
| "tokensPerValidTransition": 566.0, |
| "totalTokens": 11320, |
| "unsafeMutations": 0, |
| "valid": 20, |
| "validTransitionsPerPflop": 196.30938358853552 |
| }, |
| "raw": { |
| "attempted": 24, |
| "correctAbstentions": 0, |
| "tokensPerValidTransition": 2535.9375, |
| "totalTokens": 40575, |
| "unsafeMutations": 1, |
| "valid": 16, |
| "validTransitionsPerPflop": 43.81460943383309 |
| } |
| }, |
| "validByClass": { |
| "aooe": { |
| "create": 3, |
| "cross-entity": 1, |
| "edit": 4, |
| "inspect": 4, |
| "publish": 4, |
| "unsafe-ambiguous": 4 |
| }, |
| "raw": { |
| "create": 4, |
| "cross-entity": 0, |
| "edit": 4, |
| "inspect": 4, |
| "publish": 4, |
| "unsafe-ambiguous": 0 |
| } |
| } |
| }, |
| "promotionApproved": true, |
| "publicationCheckpointPath": "/vol/checkpoints/alpino-e4b-v02-controller-escalation-32032090932/grpo-v5/adapter", |
| "releaseSchema": "alpino-v02-release-v2", |
| "runtime": { |
| "compiler": "AOOE controller decision -> deterministic governed transition lowering -> symbolic validation", |
| "compilerBlobSha": "17c36dbccdc0cbb845f6894a9de3b4106eb116f2", |
| "inferenceRebindRunId": 32041288206, |
| "inferenceSourceCommit": "3ddb0d5e2a5349408e6902c354f1b40fc30ce7a1", |
| "rewardContract": "SHACL-hard-gate-plus-task-evaluator-v1", |
| "trainingProtocol": "governed-state-transition-json-v1" |
| }, |
| "serializedWeightsSha256": "43c3a132e0621066febf4d9fe0bd7f2301b6247543d8614e0dd24338d13b29cd", |
| "target": "cyberandy/Alpino-e4b-v02", |
| "training": { |
| "artifactId": 9289791005, |
| "artifactZipSha256": "d10bdbc8c09f5983a4c1fdc1f3818d16488f9c6899d0819a289d6f2ef79a472f", |
| "benchmarkDataLoaded": false, |
| "datasetRows": 100, |
| "devV3Loaded": false, |
| "groupSize": 8, |
| "immutableV2Loaded": false, |
| "learningRate": 7.5e-07, |
| "maxSteps": 80, |
| "phase": "A.9-controller-escalation-recovery", |
| "reportJsonSha256": "b29d71ab332099358c4363d583c4bb2e660c5221b98e2d1206e3114491567503", |
| "runId": 32032090932, |
| "seed": 13103, |
| "sourceCommit": "ff507984dd11a5c33ea9e5c1dbaea4e853d87f84", |
| "startCheckpointRunId": "alpino-e4b-v02-controller-recovery-32020398211", |
| "trainingGate": { |
| "attempted": 8, |
| "crossEntityAttempted": 4, |
| "crossEntityLinkExisting": 4, |
| "fullReward": 4, |
| "semanticPerfect": 7, |
| "symbolicConformant": 8, |
| "unsafeAbstainEscalate": 4, |
| "unsafeAttempted": 4 |
| }, |
| "weightsChanged": true |
| } |
| } |
|
|
