openpond-sdk 0.9.6 → 0.9.7
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/dist/{chunk-EMANJDJR.js → chunk-CNGJF6AM.js} +61 -198
- package/dist/chunk-CNGJF6AM.js.map +7 -0
- package/dist/chunk-SR55WVPK.js +158 -0
- package/dist/chunk-SR55WVPK.js.map +7 -0
- package/dist/dataset-marketplace.js +216 -0
- package/dist/dataset-marketplace.js.map +7 -0
- package/dist/dataset-workspaces.js +86 -2
- package/dist/dataset-workspaces.js.map +4 -4
- package/dist/experiments.js +9 -7
- package/dist/experiments.js.map +1 -1
- package/dist/index.js +9 -8
- package/dist/index.js.map +1 -1
- package/dist/types/packages/sdk/src/dataset-marketplace.d.ts +945 -0
- package/dist/types/packages/sdk/src/dataset-marketplace.d.ts.map +1 -0
- package/dist/types/packages/sdk/src/dataset-population.d.ts +97 -0
- package/dist/types/packages/sdk/src/dataset-population.d.ts.map +1 -0
- package/dist/types/packages/sdk/src/dataset-workspace-client.d.ts +146 -145
- package/dist/types/packages/sdk/src/dataset-workspace-client.d.ts.map +1 -1
- package/dist/types/packages/sdk/src/dataset-workspace-contracts.d.ts +84 -84
- package/dist/types/packages/sdk/src/dataset-workspaces.d.ts +1 -0
- package/dist/types/packages/sdk/src/dataset-workspaces.d.ts.map +1 -1
- package/dist/types/packages/sdk/src/evaluation-project-context.d.ts +1 -1
- package/dist/types/packages/sdk/src/experiment-client.d.ts +23 -23
- package/dist/types/packages/sdk/src/experiment-contracts.d.ts +3 -3
- package/dist/types/packages/sdk/src/experiment-grading-contracts.d.ts +1 -1
- package/dist/types/packages/sdk/src/experiment-scoring-contracts.d.ts +3 -3
- package/dist/types/packages/sdk/src/grader-inspection.d.ts +8 -8
- package/dist/types/packages/sdk/src/learning-batch-assets.d.ts +2 -2
- package/dist/types/packages/sdk/src/learning-client.d.ts +1 -1
- package/dist/types/packages/sdk/src/managed-training-preparation.d.ts +2 -2
- package/dist/types/packages/sdk/src/model-batch-review-preparation.d.ts +7 -7
- package/dist/types/packages/sdk/src/model-batch-review.d.ts +23 -23
- package/dist/types/packages/sdk/src/model-learning-policy.d.ts +3 -3
- package/dist/types/packages/sdk/src/model-projects.d.ts +27 -27
- package/dist/types/packages/sdk/src/model-starter-attempts.d.ts +47 -47
- package/dist/types/packages/sdk/src/model-starter-catalog.d.ts +15 -15
- package/dist/types/packages/sdk/src/model-starter-execution.d.ts +4 -4
- package/dist/types/packages/sdk/src/model-starters.d.ts +18 -18
- package/dist/types/packages/sdk/src/model-taskset-runs-contracts.d.ts +16 -16
- package/dist/types/packages/sdk/src/model-taskset-runs.d.ts +15 -15
- package/dist/types/packages/sdk/src/task-inventory.d.ts +4 -4
- package/dist/types/packages/sdk/src/taskset-authored-contracts.d.ts +46 -46
- package/dist/types/packages/sdk/src/taskset-authored-files.d.ts +30 -30
- package/dist/types/packages/sdk/src/taskset-authored-portable-release.d.ts +2 -2
- package/dist/types/packages/sdk/src/taskset-draft-api-contracts.d.ts +84 -84
- package/dist/types/packages/sdk/src/taskset-draft-client.d.ts +176 -176
- package/dist/types/packages/sdk/src/taskset-draft-core.d.ts +46 -46
- package/dist/types/packages/sdk/src/taskset-draft-dataset-artifacts.d.ts +25 -25
- package/dist/types/packages/sdk/src/taskset-draft-dataset-sources.d.ts +41 -41
- package/dist/types/packages/sdk/src/taskset-draft-document.d.ts +45 -45
- package/dist/types/packages/sdk/src/taskset-draft-package-compiler.d.ts +56 -56
- package/dist/types/packages/sdk/src/taskset-draft-source.d.ts +71 -71
- package/dist/types/packages/sdk/src/taskset-draft-workspace.d.ts +84 -84
- package/dist/types/packages/sdk/src/taskset-package-client.d.ts +1733 -1733
- package/dist/types/packages/sdk/src/taskset-package-contracts.d.ts +1083 -1083
- package/dist/types/packages/sdk/src/taskset-package-grader-projection.d.ts +2 -2
- package/dist/types/packages/sdk/src/taskset-package-learning.d.ts +18 -18
- package/dist/types/packages/sdk/src/taskset-package-projection.d.ts +59 -59
- package/dist/types/packages/sdk/src/training-acceptance.d.ts +4 -4
- package/dist/types/packages/sdk/src/training-harness-sources.d.ts +1 -1
- package/dist/types/packages/sdk/src/training-learning-batch.d.ts +68 -68
- package/dist/types/packages/sdk/src/training-project-client.d.ts +12 -12
- package/dist/types/packages/sdk/src/training-project-contracts.d.ts +21 -21
- package/dist/types/packages/sdk/src/training.d.ts +24 -24
- package/package.json +5 -1
- package/dist/chunk-EMANJDJR.js.map +0 -7
|
@@ -49,7 +49,7 @@ export declare function compileTasksetDraftWorkspace(input: {
|
|
|
49
49
|
tasks: {
|
|
50
50
|
id: string;
|
|
51
51
|
clusterKey: string;
|
|
52
|
-
split: "train" | "validation" | "
|
|
52
|
+
split: "train" | "validation" | "frozen_eval" | "test";
|
|
53
53
|
input: Record<string, unknown>;
|
|
54
54
|
expectedOutput: Record<string, unknown> | null;
|
|
55
55
|
policyVisibleContext: Record<string, unknown>;
|
|
@@ -85,7 +85,7 @@ export declare function compileTasksetDraftWorkspace(input: {
|
|
|
85
85
|
hardGate: boolean;
|
|
86
86
|
rewardEligible: boolean;
|
|
87
87
|
privileged: boolean;
|
|
88
|
-
kind: "content" | "
|
|
88
|
+
kind: "content" | "state" | "schema" | "runtime_event" | "artifact";
|
|
89
89
|
config: Record<string, unknown>;
|
|
90
90
|
} | {
|
|
91
91
|
id: string;
|
|
@@ -103,7 +103,7 @@ export declare function compileTasksetDraftWorkspace(input: {
|
|
|
103
103
|
mediaType: string;
|
|
104
104
|
visibility: "policy" | "verifier" | "host_private";
|
|
105
105
|
};
|
|
106
|
-
calibrationStatus: "
|
|
106
|
+
calibrationStatus: "passed" | "failed" | "pending";
|
|
107
107
|
model?: {
|
|
108
108
|
providerId: string;
|
|
109
109
|
modelId: string;
|
|
@@ -230,7 +230,7 @@ export declare function compileTasksetDraftWorkspace(input: {
|
|
|
230
230
|
hardGate: boolean;
|
|
231
231
|
rewardEligible: boolean;
|
|
232
232
|
privileged: boolean;
|
|
233
|
-
kind: "content" | "
|
|
233
|
+
kind: "content" | "state" | "schema" | "runtime_event" | "artifact";
|
|
234
234
|
config: Record<string, unknown>;
|
|
235
235
|
} | {
|
|
236
236
|
id: string;
|
|
@@ -248,7 +248,7 @@ export declare function compileTasksetDraftWorkspace(input: {
|
|
|
248
248
|
mediaType: string;
|
|
249
249
|
visibility: "policy" | "verifier" | "host_private";
|
|
250
250
|
};
|
|
251
|
-
calibrationStatus: "
|
|
251
|
+
calibrationStatus: "passed" | "failed" | "pending";
|
|
252
252
|
model?: {
|
|
253
253
|
providerId: string;
|
|
254
254
|
modelId: string;
|
|
@@ -417,7 +417,7 @@ export declare function compileTasksetDraftWorkspace(input: {
|
|
|
417
417
|
name: string;
|
|
418
418
|
description: string;
|
|
419
419
|
implementation: {
|
|
420
|
-
kind: "content" | "
|
|
420
|
+
kind: "content" | "state" | "schema" | "runtime_event" | "artifact";
|
|
421
421
|
config: Record<string, unknown>;
|
|
422
422
|
} | {
|
|
423
423
|
kind: "model_judge";
|
|
@@ -429,7 +429,7 @@ export declare function compileTasksetDraftWorkspace(input: {
|
|
|
429
429
|
mediaType: string;
|
|
430
430
|
visibility: "policy" | "verifier" | "host_private";
|
|
431
431
|
};
|
|
432
|
-
calibrationStatus: "
|
|
432
|
+
calibrationStatus: "passed" | "failed" | "pending";
|
|
433
433
|
model?: {
|
|
434
434
|
providerId: string;
|
|
435
435
|
modelId: string;
|
|
@@ -515,7 +515,7 @@ export declare function compileTasksetDraftWorkspace(input: {
|
|
|
515
515
|
contentHash: string;
|
|
516
516
|
revision: number;
|
|
517
517
|
};
|
|
518
|
-
role: "
|
|
518
|
+
role: "training" | "evaluation";
|
|
519
519
|
normalization: {
|
|
520
520
|
kind: "identity";
|
|
521
521
|
} | {
|
|
@@ -602,7 +602,7 @@ export declare function compileTasksetDraftWorkspace(input: {
|
|
|
602
602
|
hardGate: boolean;
|
|
603
603
|
rewardEligible: boolean;
|
|
604
604
|
privileged: boolean;
|
|
605
|
-
kind: "content" | "
|
|
605
|
+
kind: "content" | "state" | "schema" | "runtime_event" | "artifact";
|
|
606
606
|
config: Record<string, unknown>;
|
|
607
607
|
} | {
|
|
608
608
|
id: string;
|
|
@@ -620,7 +620,7 @@ export declare function compileTasksetDraftWorkspace(input: {
|
|
|
620
620
|
mediaType: string;
|
|
621
621
|
visibility: "policy" | "verifier" | "host_private";
|
|
622
622
|
};
|
|
623
|
-
calibrationStatus: "
|
|
623
|
+
calibrationStatus: "passed" | "failed" | "pending";
|
|
624
624
|
model?: {
|
|
625
625
|
providerId: string;
|
|
626
626
|
modelId: string;
|
|
@@ -769,7 +769,7 @@ export declare function compileTasksetDraftWorkspace(input: {
|
|
|
769
769
|
};
|
|
770
770
|
familyKey: string;
|
|
771
771
|
inputHash: string;
|
|
772
|
-
split: "train" | "validation" | "
|
|
772
|
+
split: "train" | "validation" | "frozen_eval" | "test";
|
|
773
773
|
}[];
|
|
774
774
|
sealedAt: string;
|
|
775
775
|
sealedBy: string;
|
|
@@ -797,7 +797,7 @@ export declare function compileTasksetDraftWorkspace(input: {
|
|
|
797
797
|
attemptId: string;
|
|
798
798
|
occurredAt: string;
|
|
799
799
|
familyKey: string | null;
|
|
800
|
-
split: "train" | "validation" | "
|
|
800
|
+
split: "train" | "validation" | "frozen_eval" | "test";
|
|
801
801
|
input: Record<string, import("zod").JSONType>;
|
|
802
802
|
observedOutput: Record<string, import("zod").JSONType> | null;
|
|
803
803
|
expected: Record<string, import("zod").JSONType> | null;
|
|
@@ -862,9 +862,9 @@ export declare function compileTasksetDraftWorkspace(input: {
|
|
|
862
862
|
} | null;
|
|
863
863
|
};
|
|
864
864
|
evidenceValidity: "pending" | "valid" | "invalid";
|
|
865
|
-
taskAdmissibility: "
|
|
866
|
-
observedQuality: "
|
|
867
|
-
targetApproval: "
|
|
865
|
+
taskAdmissibility: "pending" | "rejected" | "approved";
|
|
866
|
+
observedQuality: "passed" | "failed" | "unavailable" | "unscored";
|
|
867
|
+
targetApproval: "pending" | "rejected" | "approved" | "not_required";
|
|
868
868
|
approvedTarget: Record<string, import("zod").JSONType> | null;
|
|
869
869
|
grade: {
|
|
870
870
|
schemaVersion: "openpond.rewardComposition.v1";
|
|
@@ -882,8 +882,8 @@ export declare function compileTasksetDraftWorkspace(input: {
|
|
|
882
882
|
contentHash: string;
|
|
883
883
|
revision: number;
|
|
884
884
|
};
|
|
885
|
-
role: "
|
|
886
|
-
status: "failed" | "
|
|
885
|
+
role: "training" | "evaluation";
|
|
886
|
+
status: "failed" | "unavailable" | "scored" | "pending";
|
|
887
887
|
rawScore: number | null;
|
|
888
888
|
normalizedScore: number | null;
|
|
889
889
|
passed: boolean | null;
|
|
@@ -896,7 +896,7 @@ export declare function compileTasksetDraftWorkspace(input: {
|
|
|
896
896
|
score: number | null;
|
|
897
897
|
passed: boolean;
|
|
898
898
|
rewardEligible: boolean;
|
|
899
|
-
failureClass: "
|
|
899
|
+
failureClass: "policy_failure" | "grader_failure" | "environment_failure" | "infrastructure_failure" | "timeout" | "cancelled" | null;
|
|
900
900
|
feedback: string[];
|
|
901
901
|
visibleEvidenceRefs: string[];
|
|
902
902
|
privilegedEvidenceRefs: string[];
|
|
@@ -916,12 +916,12 @@ export declare function compileTasksetDraftWorkspace(input: {
|
|
|
916
916
|
} | undefined;
|
|
917
917
|
}[];
|
|
918
918
|
training: {
|
|
919
|
-
status: "
|
|
919
|
+
status: "scored" | "unscorable" | "not_configured";
|
|
920
920
|
score: number | null;
|
|
921
921
|
passed: boolean | null;
|
|
922
922
|
};
|
|
923
923
|
evaluation: {
|
|
924
|
-
status: "
|
|
924
|
+
status: "scored" | "unscorable" | "not_configured";
|
|
925
925
|
score: number | null;
|
|
926
926
|
passed: boolean | null;
|
|
927
927
|
};
|
|
@@ -943,8 +943,8 @@ export declare function compileTasksetDraftWorkspace(input: {
|
|
|
943
943
|
contentHash: string;
|
|
944
944
|
revision: number;
|
|
945
945
|
};
|
|
946
|
-
role: "
|
|
947
|
-
status: "failed" | "
|
|
946
|
+
role: "training" | "evaluation";
|
|
947
|
+
status: "failed" | "unavailable" | "scored" | "pending";
|
|
948
948
|
rawScore: number | null;
|
|
949
949
|
normalizedScore: number | null;
|
|
950
950
|
passed: boolean | null;
|
|
@@ -957,7 +957,7 @@ export declare function compileTasksetDraftWorkspace(input: {
|
|
|
957
957
|
score: number | null;
|
|
958
958
|
passed: boolean;
|
|
959
959
|
rewardEligible: boolean;
|
|
960
|
-
failureClass: "
|
|
960
|
+
failureClass: "policy_failure" | "grader_failure" | "environment_failure" | "infrastructure_failure" | "timeout" | "cancelled" | null;
|
|
961
961
|
feedback: string[];
|
|
962
962
|
visibleEvidenceRefs: string[];
|
|
963
963
|
privilegedEvidenceRefs: string[];
|
|
@@ -977,12 +977,12 @@ export declare function compileTasksetDraftWorkspace(input: {
|
|
|
977
977
|
} | undefined;
|
|
978
978
|
}[];
|
|
979
979
|
training: {
|
|
980
|
-
status: "
|
|
980
|
+
status: "scored" | "unscorable" | "not_configured";
|
|
981
981
|
score: number | null;
|
|
982
982
|
passed: boolean | null;
|
|
983
983
|
};
|
|
984
984
|
evaluation: {
|
|
985
|
-
status: "
|
|
985
|
+
status: "scored" | "unscorable" | "not_configured";
|
|
986
986
|
score: number | null;
|
|
987
987
|
passed: boolean | null;
|
|
988
988
|
};
|
|
@@ -997,14 +997,14 @@ export declare function compileTasksetDraftWorkspace(input: {
|
|
|
997
997
|
id: string;
|
|
998
998
|
revision: number;
|
|
999
999
|
name: string;
|
|
1000
|
-
kind: "
|
|
1000
|
+
kind: "provider" | "benchmark" | "work" | "direct" | "opentelemetry";
|
|
1001
1001
|
taskDefinition: {
|
|
1002
1002
|
id: string;
|
|
1003
1003
|
contentHash: string;
|
|
1004
1004
|
revision: number;
|
|
1005
1005
|
};
|
|
1006
1006
|
enabled: boolean;
|
|
1007
|
-
allowedSplits: ("train" | "validation" | "
|
|
1007
|
+
allowedSplits: ("train" | "validation" | "frozen_eval" | "test")[];
|
|
1008
1008
|
mapping: {
|
|
1009
1009
|
schemaVersion: "openpond.taskSourceMapping.v1";
|
|
1010
1010
|
exampleId: string;
|
|
@@ -1015,7 +1015,7 @@ export declare function compileTasksetDraftWorkspace(input: {
|
|
|
1015
1015
|
expected: string | null;
|
|
1016
1016
|
evaluatorContext: string | null;
|
|
1017
1017
|
familyKey: string;
|
|
1018
|
-
split: "train" | "validation" | "
|
|
1018
|
+
split: "train" | "validation" | "frozen_eval" | "test";
|
|
1019
1019
|
} | null;
|
|
1020
1020
|
adapterVersion: string | null;
|
|
1021
1021
|
contentHash: string;
|
|
@@ -1096,7 +1096,7 @@ export declare function compileModelTasksetDraftWorkspace(input: Parameters<type
|
|
|
1096
1096
|
tasks: {
|
|
1097
1097
|
id: string;
|
|
1098
1098
|
clusterKey: string;
|
|
1099
|
-
split: "train" | "validation" | "
|
|
1099
|
+
split: "train" | "validation" | "frozen_eval" | "test";
|
|
1100
1100
|
input: Record<string, unknown>;
|
|
1101
1101
|
expectedOutput: Record<string, unknown> | null;
|
|
1102
1102
|
policyVisibleContext: Record<string, unknown>;
|
|
@@ -1132,7 +1132,7 @@ export declare function compileModelTasksetDraftWorkspace(input: Parameters<type
|
|
|
1132
1132
|
hardGate: boolean;
|
|
1133
1133
|
rewardEligible: boolean;
|
|
1134
1134
|
privileged: boolean;
|
|
1135
|
-
kind: "content" | "
|
|
1135
|
+
kind: "content" | "state" | "schema" | "runtime_event" | "artifact";
|
|
1136
1136
|
config: Record<string, unknown>;
|
|
1137
1137
|
} | {
|
|
1138
1138
|
id: string;
|
|
@@ -1150,7 +1150,7 @@ export declare function compileModelTasksetDraftWorkspace(input: Parameters<type
|
|
|
1150
1150
|
mediaType: string;
|
|
1151
1151
|
visibility: "policy" | "verifier" | "host_private";
|
|
1152
1152
|
};
|
|
1153
|
-
calibrationStatus: "
|
|
1153
|
+
calibrationStatus: "passed" | "failed" | "pending";
|
|
1154
1154
|
model?: {
|
|
1155
1155
|
providerId: string;
|
|
1156
1156
|
modelId: string;
|
|
@@ -1277,7 +1277,7 @@ export declare function compileModelTasksetDraftWorkspace(input: Parameters<type
|
|
|
1277
1277
|
hardGate: boolean;
|
|
1278
1278
|
rewardEligible: boolean;
|
|
1279
1279
|
privileged: boolean;
|
|
1280
|
-
kind: "content" | "
|
|
1280
|
+
kind: "content" | "state" | "schema" | "runtime_event" | "artifact";
|
|
1281
1281
|
config: Record<string, unknown>;
|
|
1282
1282
|
} | {
|
|
1283
1283
|
id: string;
|
|
@@ -1295,7 +1295,7 @@ export declare function compileModelTasksetDraftWorkspace(input: Parameters<type
|
|
|
1295
1295
|
mediaType: string;
|
|
1296
1296
|
visibility: "policy" | "verifier" | "host_private";
|
|
1297
1297
|
};
|
|
1298
|
-
calibrationStatus: "
|
|
1298
|
+
calibrationStatus: "passed" | "failed" | "pending";
|
|
1299
1299
|
model?: {
|
|
1300
1300
|
providerId: string;
|
|
1301
1301
|
modelId: string;
|
|
@@ -1464,7 +1464,7 @@ export declare function compileModelTasksetDraftWorkspace(input: Parameters<type
|
|
|
1464
1464
|
name: string;
|
|
1465
1465
|
description: string;
|
|
1466
1466
|
implementation: {
|
|
1467
|
-
kind: "content" | "
|
|
1467
|
+
kind: "content" | "state" | "schema" | "runtime_event" | "artifact";
|
|
1468
1468
|
config: Record<string, unknown>;
|
|
1469
1469
|
} | {
|
|
1470
1470
|
kind: "model_judge";
|
|
@@ -1476,7 +1476,7 @@ export declare function compileModelTasksetDraftWorkspace(input: Parameters<type
|
|
|
1476
1476
|
mediaType: string;
|
|
1477
1477
|
visibility: "policy" | "verifier" | "host_private";
|
|
1478
1478
|
};
|
|
1479
|
-
calibrationStatus: "
|
|
1479
|
+
calibrationStatus: "passed" | "failed" | "pending";
|
|
1480
1480
|
model?: {
|
|
1481
1481
|
providerId: string;
|
|
1482
1482
|
modelId: string;
|
|
@@ -1562,7 +1562,7 @@ export declare function compileModelTasksetDraftWorkspace(input: Parameters<type
|
|
|
1562
1562
|
contentHash: string;
|
|
1563
1563
|
revision: number;
|
|
1564
1564
|
};
|
|
1565
|
-
role: "
|
|
1565
|
+
role: "training" | "evaluation";
|
|
1566
1566
|
normalization: {
|
|
1567
1567
|
kind: "identity";
|
|
1568
1568
|
} | {
|
|
@@ -1649,7 +1649,7 @@ export declare function compileModelTasksetDraftWorkspace(input: Parameters<type
|
|
|
1649
1649
|
hardGate: boolean;
|
|
1650
1650
|
rewardEligible: boolean;
|
|
1651
1651
|
privileged: boolean;
|
|
1652
|
-
kind: "content" | "
|
|
1652
|
+
kind: "content" | "state" | "schema" | "runtime_event" | "artifact";
|
|
1653
1653
|
config: Record<string, unknown>;
|
|
1654
1654
|
} | {
|
|
1655
1655
|
id: string;
|
|
@@ -1667,7 +1667,7 @@ export declare function compileModelTasksetDraftWorkspace(input: Parameters<type
|
|
|
1667
1667
|
mediaType: string;
|
|
1668
1668
|
visibility: "policy" | "verifier" | "host_private";
|
|
1669
1669
|
};
|
|
1670
|
-
calibrationStatus: "
|
|
1670
|
+
calibrationStatus: "passed" | "failed" | "pending";
|
|
1671
1671
|
model?: {
|
|
1672
1672
|
providerId: string;
|
|
1673
1673
|
modelId: string;
|
|
@@ -1816,7 +1816,7 @@ export declare function compileModelTasksetDraftWorkspace(input: Parameters<type
|
|
|
1816
1816
|
};
|
|
1817
1817
|
familyKey: string;
|
|
1818
1818
|
inputHash: string;
|
|
1819
|
-
split: "train" | "validation" | "
|
|
1819
|
+
split: "train" | "validation" | "frozen_eval" | "test";
|
|
1820
1820
|
}[];
|
|
1821
1821
|
sealedAt: string;
|
|
1822
1822
|
sealedBy: string;
|
|
@@ -1844,7 +1844,7 @@ export declare function compileModelTasksetDraftWorkspace(input: Parameters<type
|
|
|
1844
1844
|
attemptId: string;
|
|
1845
1845
|
occurredAt: string;
|
|
1846
1846
|
familyKey: string | null;
|
|
1847
|
-
split: "train" | "validation" | "
|
|
1847
|
+
split: "train" | "validation" | "frozen_eval" | "test";
|
|
1848
1848
|
input: Record<string, import("zod").JSONType>;
|
|
1849
1849
|
observedOutput: Record<string, import("zod").JSONType> | null;
|
|
1850
1850
|
expected: Record<string, import("zod").JSONType> | null;
|
|
@@ -1909,9 +1909,9 @@ export declare function compileModelTasksetDraftWorkspace(input: Parameters<type
|
|
|
1909
1909
|
} | null;
|
|
1910
1910
|
};
|
|
1911
1911
|
evidenceValidity: "pending" | "valid" | "invalid";
|
|
1912
|
-
taskAdmissibility: "
|
|
1913
|
-
observedQuality: "
|
|
1914
|
-
targetApproval: "
|
|
1912
|
+
taskAdmissibility: "pending" | "rejected" | "approved";
|
|
1913
|
+
observedQuality: "passed" | "failed" | "unavailable" | "unscored";
|
|
1914
|
+
targetApproval: "pending" | "rejected" | "approved" | "not_required";
|
|
1915
1915
|
approvedTarget: Record<string, import("zod").JSONType> | null;
|
|
1916
1916
|
grade: {
|
|
1917
1917
|
schemaVersion: "openpond.rewardComposition.v1";
|
|
@@ -1929,8 +1929,8 @@ export declare function compileModelTasksetDraftWorkspace(input: Parameters<type
|
|
|
1929
1929
|
contentHash: string;
|
|
1930
1930
|
revision: number;
|
|
1931
1931
|
};
|
|
1932
|
-
role: "
|
|
1933
|
-
status: "failed" | "
|
|
1932
|
+
role: "training" | "evaluation";
|
|
1933
|
+
status: "failed" | "unavailable" | "scored" | "pending";
|
|
1934
1934
|
rawScore: number | null;
|
|
1935
1935
|
normalizedScore: number | null;
|
|
1936
1936
|
passed: boolean | null;
|
|
@@ -1943,7 +1943,7 @@ export declare function compileModelTasksetDraftWorkspace(input: Parameters<type
|
|
|
1943
1943
|
score: number | null;
|
|
1944
1944
|
passed: boolean;
|
|
1945
1945
|
rewardEligible: boolean;
|
|
1946
|
-
failureClass: "
|
|
1946
|
+
failureClass: "policy_failure" | "grader_failure" | "environment_failure" | "infrastructure_failure" | "timeout" | "cancelled" | null;
|
|
1947
1947
|
feedback: string[];
|
|
1948
1948
|
visibleEvidenceRefs: string[];
|
|
1949
1949
|
privilegedEvidenceRefs: string[];
|
|
@@ -1963,12 +1963,12 @@ export declare function compileModelTasksetDraftWorkspace(input: Parameters<type
|
|
|
1963
1963
|
} | undefined;
|
|
1964
1964
|
}[];
|
|
1965
1965
|
training: {
|
|
1966
|
-
status: "
|
|
1966
|
+
status: "scored" | "unscorable" | "not_configured";
|
|
1967
1967
|
score: number | null;
|
|
1968
1968
|
passed: boolean | null;
|
|
1969
1969
|
};
|
|
1970
1970
|
evaluation: {
|
|
1971
|
-
status: "
|
|
1971
|
+
status: "scored" | "unscorable" | "not_configured";
|
|
1972
1972
|
score: number | null;
|
|
1973
1973
|
passed: boolean | null;
|
|
1974
1974
|
};
|
|
@@ -1990,8 +1990,8 @@ export declare function compileModelTasksetDraftWorkspace(input: Parameters<type
|
|
|
1990
1990
|
contentHash: string;
|
|
1991
1991
|
revision: number;
|
|
1992
1992
|
};
|
|
1993
|
-
role: "
|
|
1994
|
-
status: "failed" | "
|
|
1993
|
+
role: "training" | "evaluation";
|
|
1994
|
+
status: "failed" | "unavailable" | "scored" | "pending";
|
|
1995
1995
|
rawScore: number | null;
|
|
1996
1996
|
normalizedScore: number | null;
|
|
1997
1997
|
passed: boolean | null;
|
|
@@ -2004,7 +2004,7 @@ export declare function compileModelTasksetDraftWorkspace(input: Parameters<type
|
|
|
2004
2004
|
score: number | null;
|
|
2005
2005
|
passed: boolean;
|
|
2006
2006
|
rewardEligible: boolean;
|
|
2007
|
-
failureClass: "
|
|
2007
|
+
failureClass: "policy_failure" | "grader_failure" | "environment_failure" | "infrastructure_failure" | "timeout" | "cancelled" | null;
|
|
2008
2008
|
feedback: string[];
|
|
2009
2009
|
visibleEvidenceRefs: string[];
|
|
2010
2010
|
privilegedEvidenceRefs: string[];
|
|
@@ -2024,12 +2024,12 @@ export declare function compileModelTasksetDraftWorkspace(input: Parameters<type
|
|
|
2024
2024
|
} | undefined;
|
|
2025
2025
|
}[];
|
|
2026
2026
|
training: {
|
|
2027
|
-
status: "
|
|
2027
|
+
status: "scored" | "unscorable" | "not_configured";
|
|
2028
2028
|
score: number | null;
|
|
2029
2029
|
passed: boolean | null;
|
|
2030
2030
|
};
|
|
2031
2031
|
evaluation: {
|
|
2032
|
-
status: "
|
|
2032
|
+
status: "scored" | "unscorable" | "not_configured";
|
|
2033
2033
|
score: number | null;
|
|
2034
2034
|
passed: boolean | null;
|
|
2035
2035
|
};
|
|
@@ -2044,14 +2044,14 @@ export declare function compileModelTasksetDraftWorkspace(input: Parameters<type
|
|
|
2044
2044
|
id: string;
|
|
2045
2045
|
revision: number;
|
|
2046
2046
|
name: string;
|
|
2047
|
-
kind: "
|
|
2047
|
+
kind: "provider" | "benchmark" | "work" | "direct" | "opentelemetry";
|
|
2048
2048
|
taskDefinition: {
|
|
2049
2049
|
id: string;
|
|
2050
2050
|
contentHash: string;
|
|
2051
2051
|
revision: number;
|
|
2052
2052
|
};
|
|
2053
2053
|
enabled: boolean;
|
|
2054
|
-
allowedSplits: ("train" | "validation" | "
|
|
2054
|
+
allowedSplits: ("train" | "validation" | "frozen_eval" | "test")[];
|
|
2055
2055
|
mapping: {
|
|
2056
2056
|
schemaVersion: "openpond.taskSourceMapping.v1";
|
|
2057
2057
|
exampleId: string;
|
|
@@ -2062,7 +2062,7 @@ export declare function compileModelTasksetDraftWorkspace(input: Parameters<type
|
|
|
2062
2062
|
expected: string | null;
|
|
2063
2063
|
evaluatorContext: string | null;
|
|
2064
2064
|
familyKey: string;
|
|
2065
|
-
split: "train" | "validation" | "
|
|
2065
|
+
split: "train" | "validation" | "frozen_eval" | "test";
|
|
2066
2066
|
} | null;
|
|
2067
2067
|
adapterVersion: string | null;
|
|
2068
2068
|
contentHash: string;
|