openpond-sdk 0.4.0 → 0.5.1
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/MODEL_STARTERS.md +26 -0
- package/TRAINING_PROTOCOL.md +18 -0
- package/dist/{chunk-LLHE4BOD.js → chunk-25TBEGXB.js} +2 -2
- package/dist/{chunk-AZUCY3E5.js → chunk-6LGGLXAM.js} +3 -3
- package/dist/chunk-EPJM7VEV.js +177 -0
- package/dist/chunk-EPJM7VEV.js.map +7 -0
- package/dist/{chunk-IVEXCIHO.js → chunk-EV4PKJWN.js} +10 -10
- package/dist/{chunk-IVEXCIHO.js.map → chunk-EV4PKJWN.js.map} +1 -1
- package/dist/{chunk-ZPCK222K.js → chunk-NF63YMOL.js} +14 -7
- package/dist/{chunk-ZPCK222K.js.map → chunk-NF63YMOL.js.map} +2 -2
- package/dist/{chunk-AVW35YOH.js → chunk-ODTLCYUT.js} +151 -126
- package/dist/chunk-ODTLCYUT.js.map +7 -0
- package/dist/index.js +3 -3
- package/dist/learning.js +3 -3
- package/dist/model-starter-attempts.js +12 -162
- package/dist/model-starter-attempts.js.map +3 -3
- package/dist/model-taskset-runs.js +196 -0
- package/dist/model-taskset-runs.js.map +7 -0
- package/dist/refiner.js +1 -1
- package/dist/taskset-drafts.js +4 -4
- package/dist/taskset-packages.js +4 -4
- package/dist/training.js +3 -1
- package/dist/types/packages/sdk/src/model-taskset-runs-contracts.d.ts +568 -0
- package/dist/types/packages/sdk/src/model-taskset-runs-contracts.d.ts.map +1 -0
- package/dist/types/packages/sdk/src/model-taskset-runs.d.ts +722 -0
- package/dist/types/packages/sdk/src/model-taskset-runs.d.ts.map +1 -0
- package/dist/types/packages/sdk/src/training-grading-plan.d.ts +21 -0
- package/dist/types/packages/sdk/src/training-grading-plan.d.ts.map +1 -0
- package/dist/types/packages/sdk/src/training.d.ts +1 -0
- package/dist/types/packages/sdk/src/training.d.ts.map +1 -1
- package/package.json +7 -3
- package/dist/chunk-AVW35YOH.js.map +0 -7
- /package/dist/{chunk-LLHE4BOD.js.map → chunk-25TBEGXB.js.map} +0 -0
- /package/dist/{chunk-AZUCY3E5.js.map → chunk-6LGGLXAM.js.map} +0 -0
|
@@ -0,0 +1,568 @@
|
|
|
1
|
+
import { z } from "zod";
|
|
2
|
+
export declare const ModelTasksetRunPolicySchema: z.ZodDiscriminatedUnion<[z.ZodObject<{
|
|
3
|
+
kind: z.ZodLiteral<"hosted_chat">;
|
|
4
|
+
modelId: z.ZodString;
|
|
5
|
+
maxOutputTokens: z.ZodDefault<z.ZodNumber>;
|
|
6
|
+
temperature: z.ZodDefault<z.ZodNumber>;
|
|
7
|
+
topP: z.ZodDefault<z.ZodNumber>;
|
|
8
|
+
}, z.core.$strict>, z.ZodObject<{
|
|
9
|
+
kind: z.ZodLiteral<"fixture">;
|
|
10
|
+
}, z.core.$strict>], "kind">;
|
|
11
|
+
export declare const ModelTasksetRunRequestSchema: z.ZodObject<{
|
|
12
|
+
schemaVersion: z.ZodLiteral<"openpond.modelTasksetRunRequest.v1">;
|
|
13
|
+
operationId: z.ZodString;
|
|
14
|
+
teamId: z.ZodString;
|
|
15
|
+
modelProjectId: z.ZodString;
|
|
16
|
+
taskset: z.ZodObject<{
|
|
17
|
+
id: z.ZodString;
|
|
18
|
+
contentHash: z.ZodString;
|
|
19
|
+
revision: z.ZodNumber;
|
|
20
|
+
}, z.core.$strict>;
|
|
21
|
+
policy: z.ZodDiscriminatedUnion<[z.ZodObject<{
|
|
22
|
+
kind: z.ZodLiteral<"hosted_chat">;
|
|
23
|
+
modelId: z.ZodString;
|
|
24
|
+
maxOutputTokens: z.ZodDefault<z.ZodNumber>;
|
|
25
|
+
temperature: z.ZodDefault<z.ZodNumber>;
|
|
26
|
+
topP: z.ZodDefault<z.ZodNumber>;
|
|
27
|
+
}, z.core.$strict>, z.ZodObject<{
|
|
28
|
+
kind: z.ZodLiteral<"fixture">;
|
|
29
|
+
}, z.core.$strict>], "kind">;
|
|
30
|
+
population: z.ZodArray<z.ZodObject<{
|
|
31
|
+
receiptId: z.ZodString;
|
|
32
|
+
taskId: z.ZodString;
|
|
33
|
+
seed: z.ZodString;
|
|
34
|
+
fixtureId: z.ZodNullable<z.ZodString>;
|
|
35
|
+
}, z.core.$strict>>;
|
|
36
|
+
}, z.core.$strict>;
|
|
37
|
+
export type ModelTasksetRunRequest = z.infer<typeof ModelTasksetRunRequestSchema>;
|
|
38
|
+
export declare const ModelTasksetRunSummarySchema: z.ZodObject<{
|
|
39
|
+
schemaVersion: z.ZodLiteral<"openpond.modelTasksetRunSummary.v1">;
|
|
40
|
+
id: z.ZodString;
|
|
41
|
+
revision: z.ZodNumber;
|
|
42
|
+
teamId: z.ZodString;
|
|
43
|
+
modelProjectId: z.ZodString;
|
|
44
|
+
operationId: z.ZodString;
|
|
45
|
+
taskset: z.ZodObject<{
|
|
46
|
+
id: z.ZodString;
|
|
47
|
+
contentHash: z.ZodString;
|
|
48
|
+
revision: z.ZodNumber;
|
|
49
|
+
}, z.core.$strict>;
|
|
50
|
+
policyKind: z.ZodEnum<{
|
|
51
|
+
fixture: "fixture";
|
|
52
|
+
hosted_chat: "hosted_chat";
|
|
53
|
+
}>;
|
|
54
|
+
manifestHash: z.ZodString;
|
|
55
|
+
status: z.ZodEnum<{
|
|
56
|
+
running: "running";
|
|
57
|
+
failed: "failed";
|
|
58
|
+
queued: "queued";
|
|
59
|
+
completed: "completed";
|
|
60
|
+
cancelled: "cancelled";
|
|
61
|
+
cancelling: "cancelling";
|
|
62
|
+
}>;
|
|
63
|
+
totalCount: z.ZodNumber;
|
|
64
|
+
counts: z.ZodObject<{
|
|
65
|
+
pending: z.ZodNumber;
|
|
66
|
+
running: z.ZodNumber;
|
|
67
|
+
completed: z.ZodNumber;
|
|
68
|
+
failed: z.ZodNumber;
|
|
69
|
+
cancelled: z.ZodNumber;
|
|
70
|
+
}, z.core.$strict>;
|
|
71
|
+
createdAt: z.ZodISODateTime;
|
|
72
|
+
startedAt: z.ZodNullable<z.ZodISODateTime>;
|
|
73
|
+
completedAt: z.ZodNullable<z.ZodISODateTime>;
|
|
74
|
+
cleanupComplete: z.ZodBoolean;
|
|
75
|
+
resultAvailable: z.ZodBoolean;
|
|
76
|
+
score: z.ZodNullable<z.ZodNumber>;
|
|
77
|
+
metricName: z.ZodString;
|
|
78
|
+
error: z.ZodNullable<z.ZodObject<{
|
|
79
|
+
code: z.ZodString;
|
|
80
|
+
message: z.ZodString;
|
|
81
|
+
}, z.core.$strict>>;
|
|
82
|
+
}, z.core.$strict>;
|
|
83
|
+
export type ModelTasksetRunSummary = z.infer<typeof ModelTasksetRunSummarySchema>;
|
|
84
|
+
export declare const ModelTasksetRunDetailsSchema: z.ZodObject<{
|
|
85
|
+
summary: z.ZodObject<{
|
|
86
|
+
schemaVersion: z.ZodLiteral<"openpond.modelTasksetRunSummary.v1">;
|
|
87
|
+
id: z.ZodString;
|
|
88
|
+
revision: z.ZodNumber;
|
|
89
|
+
teamId: z.ZodString;
|
|
90
|
+
modelProjectId: z.ZodString;
|
|
91
|
+
operationId: z.ZodString;
|
|
92
|
+
taskset: z.ZodObject<{
|
|
93
|
+
id: z.ZodString;
|
|
94
|
+
contentHash: z.ZodString;
|
|
95
|
+
revision: z.ZodNumber;
|
|
96
|
+
}, z.core.$strict>;
|
|
97
|
+
policyKind: z.ZodEnum<{
|
|
98
|
+
fixture: "fixture";
|
|
99
|
+
hosted_chat: "hosted_chat";
|
|
100
|
+
}>;
|
|
101
|
+
manifestHash: z.ZodString;
|
|
102
|
+
status: z.ZodEnum<{
|
|
103
|
+
running: "running";
|
|
104
|
+
failed: "failed";
|
|
105
|
+
queued: "queued";
|
|
106
|
+
completed: "completed";
|
|
107
|
+
cancelled: "cancelled";
|
|
108
|
+
cancelling: "cancelling";
|
|
109
|
+
}>;
|
|
110
|
+
totalCount: z.ZodNumber;
|
|
111
|
+
counts: z.ZodObject<{
|
|
112
|
+
pending: z.ZodNumber;
|
|
113
|
+
running: z.ZodNumber;
|
|
114
|
+
completed: z.ZodNumber;
|
|
115
|
+
failed: z.ZodNumber;
|
|
116
|
+
cancelled: z.ZodNumber;
|
|
117
|
+
}, z.core.$strict>;
|
|
118
|
+
createdAt: z.ZodISODateTime;
|
|
119
|
+
startedAt: z.ZodNullable<z.ZodISODateTime>;
|
|
120
|
+
completedAt: z.ZodNullable<z.ZodISODateTime>;
|
|
121
|
+
cleanupComplete: z.ZodBoolean;
|
|
122
|
+
resultAvailable: z.ZodBoolean;
|
|
123
|
+
score: z.ZodNullable<z.ZodNumber>;
|
|
124
|
+
metricName: z.ZodString;
|
|
125
|
+
error: z.ZodNullable<z.ZodObject<{
|
|
126
|
+
code: z.ZodString;
|
|
127
|
+
message: z.ZodString;
|
|
128
|
+
}, z.core.$strict>>;
|
|
129
|
+
}, z.core.$strict>;
|
|
130
|
+
request: z.ZodObject<{
|
|
131
|
+
schemaVersion: z.ZodLiteral<"openpond.modelTasksetRunRequest.v1">;
|
|
132
|
+
operationId: z.ZodString;
|
|
133
|
+
teamId: z.ZodString;
|
|
134
|
+
modelProjectId: z.ZodString;
|
|
135
|
+
taskset: z.ZodObject<{
|
|
136
|
+
id: z.ZodString;
|
|
137
|
+
contentHash: z.ZodString;
|
|
138
|
+
revision: z.ZodNumber;
|
|
139
|
+
}, z.core.$strict>;
|
|
140
|
+
policy: z.ZodDiscriminatedUnion<[z.ZodObject<{
|
|
141
|
+
kind: z.ZodLiteral<"hosted_chat">;
|
|
142
|
+
modelId: z.ZodString;
|
|
143
|
+
maxOutputTokens: z.ZodDefault<z.ZodNumber>;
|
|
144
|
+
temperature: z.ZodDefault<z.ZodNumber>;
|
|
145
|
+
topP: z.ZodDefault<z.ZodNumber>;
|
|
146
|
+
}, z.core.$strict>, z.ZodObject<{
|
|
147
|
+
kind: z.ZodLiteral<"fixture">;
|
|
148
|
+
}, z.core.$strict>], "kind">;
|
|
149
|
+
population: z.ZodArray<z.ZodObject<{
|
|
150
|
+
receiptId: z.ZodString;
|
|
151
|
+
taskId: z.ZodString;
|
|
152
|
+
seed: z.ZodString;
|
|
153
|
+
fixtureId: z.ZodNullable<z.ZodString>;
|
|
154
|
+
}, z.core.$strict>>;
|
|
155
|
+
}, z.core.$strict>;
|
|
156
|
+
manifest: z.ZodObject<{
|
|
157
|
+
schemaVersion: z.ZodLiteral<"openpond.tasksetRunManifest.v1">;
|
|
158
|
+
id: z.ZodString;
|
|
159
|
+
tasksetRelease: z.ZodObject<{
|
|
160
|
+
id: z.ZodString;
|
|
161
|
+
contentHash: z.ZodString;
|
|
162
|
+
}, z.core.$strict>;
|
|
163
|
+
packageHash: z.ZodString;
|
|
164
|
+
execution: z.ZodDiscriminatedUnion<[z.ZodObject<{
|
|
165
|
+
kind: z.ZodLiteral<"taskset">;
|
|
166
|
+
environmentRelease: z.ZodObject<{
|
|
167
|
+
id: z.ZodString;
|
|
168
|
+
contentHash: z.ZodString;
|
|
169
|
+
}, z.core.$strict>;
|
|
170
|
+
verifierSetRelease: z.ZodObject<{
|
|
171
|
+
id: z.ZodString;
|
|
172
|
+
contentHash: z.ZodString;
|
|
173
|
+
}, z.core.$strict>;
|
|
174
|
+
policyHash: z.ZodString;
|
|
175
|
+
}, z.core.$strict>, z.ZodObject<{
|
|
176
|
+
kind: z.ZodLiteral<"harness">;
|
|
177
|
+
harnessRelease: z.ZodObject<{
|
|
178
|
+
id: z.ZodString;
|
|
179
|
+
contentHash: z.ZodString;
|
|
180
|
+
}, z.core.$strict>;
|
|
181
|
+
}, z.core.$strict>], "kind">;
|
|
182
|
+
policy: z.ZodDiscriminatedUnion<[z.ZodObject<{
|
|
183
|
+
kind: z.ZodLiteral<"model">;
|
|
184
|
+
model: z.ZodObject<{
|
|
185
|
+
provider: z.ZodString;
|
|
186
|
+
model: z.ZodString;
|
|
187
|
+
revision: z.ZodDefault<z.ZodNullable<z.ZodString>>;
|
|
188
|
+
artifactHash: z.ZodDefault<z.ZodNullable<z.ZodString>>;
|
|
189
|
+
tokenizerRevision: z.ZodDefault<z.ZodNullable<z.ZodString>>;
|
|
190
|
+
chatTemplateHash: z.ZodDefault<z.ZodNullable<z.ZodString>>;
|
|
191
|
+
}, z.core.$strict>;
|
|
192
|
+
configurationHash: z.ZodString;
|
|
193
|
+
}, z.core.$strict>, z.ZodObject<{
|
|
194
|
+
kind: z.ZodLiteral<"fixture">;
|
|
195
|
+
}, z.core.$strict>], "kind">;
|
|
196
|
+
gradingRole: z.ZodLiteral<"evaluation">;
|
|
197
|
+
metricPolicy: z.ZodObject<{
|
|
198
|
+
schemaVersion: z.ZodLiteral<"openpond.tasksetMetricPolicy.v1">;
|
|
199
|
+
primaryMetric: z.ZodString;
|
|
200
|
+
aggregation: z.ZodEnum<{
|
|
201
|
+
mean_score: "mean_score";
|
|
202
|
+
pass_rate: "pass_rate";
|
|
203
|
+
weighted_mean: "weighted_mean";
|
|
204
|
+
custom: "custom";
|
|
205
|
+
}>;
|
|
206
|
+
missingReward: z.ZodEnum<{
|
|
207
|
+
zero: "zero";
|
|
208
|
+
exclude: "exclude";
|
|
209
|
+
}>;
|
|
210
|
+
taskWeights: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodNumber>>;
|
|
211
|
+
customAggregator: z.ZodNullable<z.ZodObject<{
|
|
212
|
+
module: z.ZodString;
|
|
213
|
+
exportName: z.ZodString;
|
|
214
|
+
contentHash: z.ZodString;
|
|
215
|
+
timeoutMs: z.ZodNumber;
|
|
216
|
+
networkPolicy: z.ZodLiteral<"none">;
|
|
217
|
+
}, z.core.$strict>>;
|
|
218
|
+
}, z.core.$strict>;
|
|
219
|
+
population: z.ZodArray<z.ZodObject<{
|
|
220
|
+
receiptId: z.ZodString;
|
|
221
|
+
taskId: z.ZodString;
|
|
222
|
+
seed: z.ZodString;
|
|
223
|
+
fixtureId: z.ZodNullable<z.ZodString>;
|
|
224
|
+
}, z.core.$strict>>;
|
|
225
|
+
runtimeTarget: z.ZodObject<{
|
|
226
|
+
adapterId: z.ZodString;
|
|
227
|
+
placement: z.ZodEnum<{
|
|
228
|
+
local: "local";
|
|
229
|
+
remote: "remote";
|
|
230
|
+
colocated: "colocated";
|
|
231
|
+
provider_native: "provider_native";
|
|
232
|
+
}>;
|
|
233
|
+
runtimeVersion: z.ZodString;
|
|
234
|
+
capabilityReceipt: z.ZodString;
|
|
235
|
+
}, z.core.$strict>;
|
|
236
|
+
limits: z.ZodObject<{
|
|
237
|
+
maxTurns: z.ZodNumber;
|
|
238
|
+
timeoutMs: z.ZodNumber;
|
|
239
|
+
maxOutputBytes: z.ZodNumber;
|
|
240
|
+
maximumSpendUsd: z.ZodNullable<z.ZodNumber>;
|
|
241
|
+
}, z.core.$strict>;
|
|
242
|
+
createdAt: z.ZodString;
|
|
243
|
+
metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
|
|
244
|
+
contentHash: z.ZodString;
|
|
245
|
+
}, z.core.$strict>;
|
|
246
|
+
policySnapshot: z.ZodNullable<z.ZodObject<{
|
|
247
|
+
modelId: z.ZodString;
|
|
248
|
+
provider: z.ZodString;
|
|
249
|
+
upstreamModelId: z.ZodString;
|
|
250
|
+
configurationHash: z.ZodString;
|
|
251
|
+
}, z.core.$strict>>;
|
|
252
|
+
}, z.core.$strict>;
|
|
253
|
+
export type ModelTasksetRunDetails = z.infer<typeof ModelTasksetRunDetailsSchema>;
|
|
254
|
+
export declare const ModelTasksetRunListQuerySchema: z.ZodObject<{
|
|
255
|
+
modelProjectId: z.ZodString;
|
|
256
|
+
afterId: z.ZodOptional<z.ZodString>;
|
|
257
|
+
limit: z.ZodDefault<z.ZodNumber>;
|
|
258
|
+
}, z.core.$strict>;
|
|
259
|
+
export declare const ModelTasksetRunPageSchema: z.ZodObject<{
|
|
260
|
+
items: z.ZodArray<z.ZodObject<{
|
|
261
|
+
schemaVersion: z.ZodLiteral<"openpond.modelTasksetRunSummary.v1">;
|
|
262
|
+
id: z.ZodString;
|
|
263
|
+
revision: z.ZodNumber;
|
|
264
|
+
teamId: z.ZodString;
|
|
265
|
+
modelProjectId: z.ZodString;
|
|
266
|
+
operationId: z.ZodString;
|
|
267
|
+
taskset: z.ZodObject<{
|
|
268
|
+
id: z.ZodString;
|
|
269
|
+
contentHash: z.ZodString;
|
|
270
|
+
revision: z.ZodNumber;
|
|
271
|
+
}, z.core.$strict>;
|
|
272
|
+
policyKind: z.ZodEnum<{
|
|
273
|
+
fixture: "fixture";
|
|
274
|
+
hosted_chat: "hosted_chat";
|
|
275
|
+
}>;
|
|
276
|
+
manifestHash: z.ZodString;
|
|
277
|
+
status: z.ZodEnum<{
|
|
278
|
+
running: "running";
|
|
279
|
+
failed: "failed";
|
|
280
|
+
queued: "queued";
|
|
281
|
+
completed: "completed";
|
|
282
|
+
cancelled: "cancelled";
|
|
283
|
+
cancelling: "cancelling";
|
|
284
|
+
}>;
|
|
285
|
+
totalCount: z.ZodNumber;
|
|
286
|
+
counts: z.ZodObject<{
|
|
287
|
+
pending: z.ZodNumber;
|
|
288
|
+
running: z.ZodNumber;
|
|
289
|
+
completed: z.ZodNumber;
|
|
290
|
+
failed: z.ZodNumber;
|
|
291
|
+
cancelled: z.ZodNumber;
|
|
292
|
+
}, z.core.$strict>;
|
|
293
|
+
createdAt: z.ZodISODateTime;
|
|
294
|
+
startedAt: z.ZodNullable<z.ZodISODateTime>;
|
|
295
|
+
completedAt: z.ZodNullable<z.ZodISODateTime>;
|
|
296
|
+
cleanupComplete: z.ZodBoolean;
|
|
297
|
+
resultAvailable: z.ZodBoolean;
|
|
298
|
+
score: z.ZodNullable<z.ZodNumber>;
|
|
299
|
+
metricName: z.ZodString;
|
|
300
|
+
error: z.ZodNullable<z.ZodObject<{
|
|
301
|
+
code: z.ZodString;
|
|
302
|
+
message: z.ZodString;
|
|
303
|
+
}, z.core.$strict>>;
|
|
304
|
+
}, z.core.$strict>>;
|
|
305
|
+
nextCursor: z.ZodNullable<z.ZodString>;
|
|
306
|
+
}, z.core.$strict>;
|
|
307
|
+
export declare const ModelTasksetRunResultSchema: z.ZodObject<{
|
|
308
|
+
schemaVersion: z.ZodLiteral<"openpond.modelTasksetRunResult.v1">;
|
|
309
|
+
run: z.ZodObject<{
|
|
310
|
+
summary: z.ZodObject<{
|
|
311
|
+
schemaVersion: z.ZodLiteral<"openpond.modelTasksetRunSummary.v1">;
|
|
312
|
+
id: z.ZodString;
|
|
313
|
+
revision: z.ZodNumber;
|
|
314
|
+
teamId: z.ZodString;
|
|
315
|
+
modelProjectId: z.ZodString;
|
|
316
|
+
operationId: z.ZodString;
|
|
317
|
+
taskset: z.ZodObject<{
|
|
318
|
+
id: z.ZodString;
|
|
319
|
+
contentHash: z.ZodString;
|
|
320
|
+
revision: z.ZodNumber;
|
|
321
|
+
}, z.core.$strict>;
|
|
322
|
+
policyKind: z.ZodEnum<{
|
|
323
|
+
fixture: "fixture";
|
|
324
|
+
hosted_chat: "hosted_chat";
|
|
325
|
+
}>;
|
|
326
|
+
manifestHash: z.ZodString;
|
|
327
|
+
status: z.ZodEnum<{
|
|
328
|
+
running: "running";
|
|
329
|
+
failed: "failed";
|
|
330
|
+
queued: "queued";
|
|
331
|
+
completed: "completed";
|
|
332
|
+
cancelled: "cancelled";
|
|
333
|
+
cancelling: "cancelling";
|
|
334
|
+
}>;
|
|
335
|
+
totalCount: z.ZodNumber;
|
|
336
|
+
counts: z.ZodObject<{
|
|
337
|
+
pending: z.ZodNumber;
|
|
338
|
+
running: z.ZodNumber;
|
|
339
|
+
completed: z.ZodNumber;
|
|
340
|
+
failed: z.ZodNumber;
|
|
341
|
+
cancelled: z.ZodNumber;
|
|
342
|
+
}, z.core.$strict>;
|
|
343
|
+
createdAt: z.ZodISODateTime;
|
|
344
|
+
startedAt: z.ZodNullable<z.ZodISODateTime>;
|
|
345
|
+
completedAt: z.ZodNullable<z.ZodISODateTime>;
|
|
346
|
+
cleanupComplete: z.ZodBoolean;
|
|
347
|
+
resultAvailable: z.ZodBoolean;
|
|
348
|
+
score: z.ZodNullable<z.ZodNumber>;
|
|
349
|
+
metricName: z.ZodString;
|
|
350
|
+
error: z.ZodNullable<z.ZodObject<{
|
|
351
|
+
code: z.ZodString;
|
|
352
|
+
message: z.ZodString;
|
|
353
|
+
}, z.core.$strict>>;
|
|
354
|
+
}, z.core.$strict>;
|
|
355
|
+
request: z.ZodObject<{
|
|
356
|
+
schemaVersion: z.ZodLiteral<"openpond.modelTasksetRunRequest.v1">;
|
|
357
|
+
operationId: z.ZodString;
|
|
358
|
+
teamId: z.ZodString;
|
|
359
|
+
modelProjectId: z.ZodString;
|
|
360
|
+
taskset: z.ZodObject<{
|
|
361
|
+
id: z.ZodString;
|
|
362
|
+
contentHash: z.ZodString;
|
|
363
|
+
revision: z.ZodNumber;
|
|
364
|
+
}, z.core.$strict>;
|
|
365
|
+
policy: z.ZodDiscriminatedUnion<[z.ZodObject<{
|
|
366
|
+
kind: z.ZodLiteral<"hosted_chat">;
|
|
367
|
+
modelId: z.ZodString;
|
|
368
|
+
maxOutputTokens: z.ZodDefault<z.ZodNumber>;
|
|
369
|
+
temperature: z.ZodDefault<z.ZodNumber>;
|
|
370
|
+
topP: z.ZodDefault<z.ZodNumber>;
|
|
371
|
+
}, z.core.$strict>, z.ZodObject<{
|
|
372
|
+
kind: z.ZodLiteral<"fixture">;
|
|
373
|
+
}, z.core.$strict>], "kind">;
|
|
374
|
+
population: z.ZodArray<z.ZodObject<{
|
|
375
|
+
receiptId: z.ZodString;
|
|
376
|
+
taskId: z.ZodString;
|
|
377
|
+
seed: z.ZodString;
|
|
378
|
+
fixtureId: z.ZodNullable<z.ZodString>;
|
|
379
|
+
}, z.core.$strict>>;
|
|
380
|
+
}, z.core.$strict>;
|
|
381
|
+
manifest: z.ZodObject<{
|
|
382
|
+
schemaVersion: z.ZodLiteral<"openpond.tasksetRunManifest.v1">;
|
|
383
|
+
id: z.ZodString;
|
|
384
|
+
tasksetRelease: z.ZodObject<{
|
|
385
|
+
id: z.ZodString;
|
|
386
|
+
contentHash: z.ZodString;
|
|
387
|
+
}, z.core.$strict>;
|
|
388
|
+
packageHash: z.ZodString;
|
|
389
|
+
execution: z.ZodDiscriminatedUnion<[z.ZodObject<{
|
|
390
|
+
kind: z.ZodLiteral<"taskset">;
|
|
391
|
+
environmentRelease: z.ZodObject<{
|
|
392
|
+
id: z.ZodString;
|
|
393
|
+
contentHash: z.ZodString;
|
|
394
|
+
}, z.core.$strict>;
|
|
395
|
+
verifierSetRelease: z.ZodObject<{
|
|
396
|
+
id: z.ZodString;
|
|
397
|
+
contentHash: z.ZodString;
|
|
398
|
+
}, z.core.$strict>;
|
|
399
|
+
policyHash: z.ZodString;
|
|
400
|
+
}, z.core.$strict>, z.ZodObject<{
|
|
401
|
+
kind: z.ZodLiteral<"harness">;
|
|
402
|
+
harnessRelease: z.ZodObject<{
|
|
403
|
+
id: z.ZodString;
|
|
404
|
+
contentHash: z.ZodString;
|
|
405
|
+
}, z.core.$strict>;
|
|
406
|
+
}, z.core.$strict>], "kind">;
|
|
407
|
+
policy: z.ZodDiscriminatedUnion<[z.ZodObject<{
|
|
408
|
+
kind: z.ZodLiteral<"model">;
|
|
409
|
+
model: z.ZodObject<{
|
|
410
|
+
provider: z.ZodString;
|
|
411
|
+
model: z.ZodString;
|
|
412
|
+
revision: z.ZodDefault<z.ZodNullable<z.ZodString>>;
|
|
413
|
+
artifactHash: z.ZodDefault<z.ZodNullable<z.ZodString>>;
|
|
414
|
+
tokenizerRevision: z.ZodDefault<z.ZodNullable<z.ZodString>>;
|
|
415
|
+
chatTemplateHash: z.ZodDefault<z.ZodNullable<z.ZodString>>;
|
|
416
|
+
}, z.core.$strict>;
|
|
417
|
+
configurationHash: z.ZodString;
|
|
418
|
+
}, z.core.$strict>, z.ZodObject<{
|
|
419
|
+
kind: z.ZodLiteral<"fixture">;
|
|
420
|
+
}, z.core.$strict>], "kind">;
|
|
421
|
+
gradingRole: z.ZodLiteral<"evaluation">;
|
|
422
|
+
metricPolicy: z.ZodObject<{
|
|
423
|
+
schemaVersion: z.ZodLiteral<"openpond.tasksetMetricPolicy.v1">;
|
|
424
|
+
primaryMetric: z.ZodString;
|
|
425
|
+
aggregation: z.ZodEnum<{
|
|
426
|
+
mean_score: "mean_score";
|
|
427
|
+
pass_rate: "pass_rate";
|
|
428
|
+
weighted_mean: "weighted_mean";
|
|
429
|
+
custom: "custom";
|
|
430
|
+
}>;
|
|
431
|
+
missingReward: z.ZodEnum<{
|
|
432
|
+
zero: "zero";
|
|
433
|
+
exclude: "exclude";
|
|
434
|
+
}>;
|
|
435
|
+
taskWeights: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodNumber>>;
|
|
436
|
+
customAggregator: z.ZodNullable<z.ZodObject<{
|
|
437
|
+
module: z.ZodString;
|
|
438
|
+
exportName: z.ZodString;
|
|
439
|
+
contentHash: z.ZodString;
|
|
440
|
+
timeoutMs: z.ZodNumber;
|
|
441
|
+
networkPolicy: z.ZodLiteral<"none">;
|
|
442
|
+
}, z.core.$strict>>;
|
|
443
|
+
}, z.core.$strict>;
|
|
444
|
+
population: z.ZodArray<z.ZodObject<{
|
|
445
|
+
receiptId: z.ZodString;
|
|
446
|
+
taskId: z.ZodString;
|
|
447
|
+
seed: z.ZodString;
|
|
448
|
+
fixtureId: z.ZodNullable<z.ZodString>;
|
|
449
|
+
}, z.core.$strict>>;
|
|
450
|
+
runtimeTarget: z.ZodObject<{
|
|
451
|
+
adapterId: z.ZodString;
|
|
452
|
+
placement: z.ZodEnum<{
|
|
453
|
+
local: "local";
|
|
454
|
+
remote: "remote";
|
|
455
|
+
colocated: "colocated";
|
|
456
|
+
provider_native: "provider_native";
|
|
457
|
+
}>;
|
|
458
|
+
runtimeVersion: z.ZodString;
|
|
459
|
+
capabilityReceipt: z.ZodString;
|
|
460
|
+
}, z.core.$strict>;
|
|
461
|
+
limits: z.ZodObject<{
|
|
462
|
+
maxTurns: z.ZodNumber;
|
|
463
|
+
timeoutMs: z.ZodNumber;
|
|
464
|
+
maxOutputBytes: z.ZodNumber;
|
|
465
|
+
maximumSpendUsd: z.ZodNullable<z.ZodNumber>;
|
|
466
|
+
}, z.core.$strict>;
|
|
467
|
+
createdAt: z.ZodString;
|
|
468
|
+
metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
|
|
469
|
+
contentHash: z.ZodString;
|
|
470
|
+
}, z.core.$strict>;
|
|
471
|
+
policySnapshot: z.ZodNullable<z.ZodObject<{
|
|
472
|
+
modelId: z.ZodString;
|
|
473
|
+
provider: z.ZodString;
|
|
474
|
+
upstreamModelId: z.ZodString;
|
|
475
|
+
configurationHash: z.ZodString;
|
|
476
|
+
}, z.core.$strict>>;
|
|
477
|
+
}, z.core.$strict>;
|
|
478
|
+
receipts: z.ZodArray<z.ZodObject<{
|
|
479
|
+
schemaVersion: z.ZodLiteral<"openpond.attemptReceipt.v1">;
|
|
480
|
+
id: z.ZodString;
|
|
481
|
+
runManifest: z.ZodObject<{
|
|
482
|
+
id: z.ZodString;
|
|
483
|
+
contentHash: z.ZodString;
|
|
484
|
+
}, z.core.$strict>;
|
|
485
|
+
taskId: z.ZodString;
|
|
486
|
+
seed: z.ZodString;
|
|
487
|
+
terminal: z.ZodBoolean;
|
|
488
|
+
failureClass: z.ZodNullable<z.ZodEnum<{
|
|
489
|
+
policy_failure: "policy_failure";
|
|
490
|
+
grader_failure: "grader_failure";
|
|
491
|
+
environment_failure: "environment_failure";
|
|
492
|
+
infrastructure_failure: "infrastructure_failure";
|
|
493
|
+
timeout: "timeout";
|
|
494
|
+
cancelled: "cancelled";
|
|
495
|
+
}>>;
|
|
496
|
+
outputHash: z.ZodNullable<z.ZodString>;
|
|
497
|
+
traceHash: z.ZodString;
|
|
498
|
+
artifactRefs: z.ZodArray<z.ZodObject<{
|
|
499
|
+
id: z.ZodString;
|
|
500
|
+
contentHash: z.ZodString;
|
|
501
|
+
mediaType: z.ZodDefault<z.ZodNullable<z.ZodString>>;
|
|
502
|
+
sizeBytes: z.ZodDefault<z.ZodNullable<z.ZodNumber>>;
|
|
503
|
+
}, z.core.$strict>>;
|
|
504
|
+
graderEvidenceRefs: z.ZodArray<z.ZodObject<{
|
|
505
|
+
id: z.ZodString;
|
|
506
|
+
contentHash: z.ZodString;
|
|
507
|
+
mediaType: z.ZodDefault<z.ZodNullable<z.ZodString>>;
|
|
508
|
+
sizeBytes: z.ZodDefault<z.ZodNullable<z.ZodNumber>>;
|
|
509
|
+
}, z.core.$strict>>;
|
|
510
|
+
startedAt: z.ZodString;
|
|
511
|
+
completedAt: z.ZodString;
|
|
512
|
+
latencyMs: z.ZodNumber;
|
|
513
|
+
costUsd: z.ZodNullable<z.ZodNumber>;
|
|
514
|
+
legacyAttemptRef: z.ZodDefault<z.ZodNullable<z.ZodString>>;
|
|
515
|
+
metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
|
|
516
|
+
contentHash: z.ZodString;
|
|
517
|
+
}, z.core.$strict>>;
|
|
518
|
+
metric: z.ZodNullable<z.ZodObject<{
|
|
519
|
+
schemaVersion: z.ZodLiteral<"openpond.tasksetMetricResult.v1">;
|
|
520
|
+
runManifest: z.ZodObject<{
|
|
521
|
+
id: z.ZodString;
|
|
522
|
+
contentHash: z.ZodString;
|
|
523
|
+
}, z.core.$strict>;
|
|
524
|
+
tasksetRelease: z.ZodObject<{
|
|
525
|
+
id: z.ZodString;
|
|
526
|
+
contentHash: z.ZodString;
|
|
527
|
+
}, z.core.$strict>;
|
|
528
|
+
policy: z.ZodObject<{
|
|
529
|
+
schemaVersion: z.ZodLiteral<"openpond.tasksetMetricPolicy.v1">;
|
|
530
|
+
primaryMetric: z.ZodString;
|
|
531
|
+
aggregation: z.ZodEnum<{
|
|
532
|
+
mean_score: "mean_score";
|
|
533
|
+
pass_rate: "pass_rate";
|
|
534
|
+
weighted_mean: "weighted_mean";
|
|
535
|
+
custom: "custom";
|
|
536
|
+
}>;
|
|
537
|
+
missingReward: z.ZodEnum<{
|
|
538
|
+
zero: "zero";
|
|
539
|
+
exclude: "exclude";
|
|
540
|
+
}>;
|
|
541
|
+
taskWeights: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodNumber>>;
|
|
542
|
+
customAggregator: z.ZodNullable<z.ZodObject<{
|
|
543
|
+
module: z.ZodString;
|
|
544
|
+
exportName: z.ZodString;
|
|
545
|
+
contentHash: z.ZodString;
|
|
546
|
+
timeoutMs: z.ZodNumber;
|
|
547
|
+
networkPolicy: z.ZodLiteral<"none">;
|
|
548
|
+
}, z.core.$strict>>;
|
|
549
|
+
}, z.core.$strict>;
|
|
550
|
+
policyHash: z.ZodString;
|
|
551
|
+
receiptRefs: z.ZodArray<z.ZodObject<{
|
|
552
|
+
id: z.ZodString;
|
|
553
|
+
contentHash: z.ZodString;
|
|
554
|
+
}, z.core.$strict>>;
|
|
555
|
+
includedCount: z.ZodNumber;
|
|
556
|
+
missingRewardCount: z.ZodNumber;
|
|
557
|
+
excludedCount: z.ZodNumber;
|
|
558
|
+
value: z.ZodNullable<z.ZodNumber>;
|
|
559
|
+
contentHash: z.ZodString;
|
|
560
|
+
}, z.core.$strict>>;
|
|
561
|
+
contentHash: z.ZodString;
|
|
562
|
+
}, z.core.$strict>;
|
|
563
|
+
export type ModelTasksetRunResult = z.infer<typeof ModelTasksetRunResultSchema>;
|
|
564
|
+
/** Server artifact ownership authenticates the producer. These checks bind all
|
|
565
|
+
* portable projections to the admitted request, population and policy. */
|
|
566
|
+
export declare function verifyModelTasksetRunDetails(value: unknown): Promise<ModelTasksetRunDetails>;
|
|
567
|
+
export declare function verifyModelTasksetRunResult(value: unknown): Promise<ModelTasksetRunResult>;
|
|
568
|
+
//# sourceMappingURL=model-taskset-runs-contracts.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"model-taskset-runs-contracts.d.ts","sourceRoot":"","sources":["../../../../../src/model-taskset-runs-contracts.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAUxB,eAAO,MAAM,2BAA2B;;;;;;;;4BAGtC,CAAC;AACH,eAAO,MAAM,4BAA4B;;;;;;;;;;;;;;;;;;;;;;;;;kBASvC,CAAC;AACH,MAAM,MAAM,sBAAsB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,4BAA4B,CAAC,CAAC;AAClF,eAAO,MAAM,4BAA4B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAmBvC,CAAC;AACH,MAAM,MAAM,sBAAsB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,4BAA4B,CAAC,CAAC;AAClF,eAAO,MAAM,4BAA4B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAK9B,CAAC;AACZ,MAAM,MAAM,sBAAsB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,4BAA4B,CAAC,CAAC;AAClF,eAAO,MAAM,8BAA8B;;;;kBAAqI,CAAC;AACjL,eAAO,MAAM,yBAAyB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAAgH,CAAC;AACvJ,eAAO,MAAM,2BAA2B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAM7B,CAAC;AACZ,MAAM,MAAM,qBAAqB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,2BAA2B,CAAC,CAAC;AAEhF;0EAC0E;AAC1E,wBAAsB,4BAA4B,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO,CAAC,sBAAsB,CAAC,CAkBlG;AAED,wBAAsB,2BAA2B,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO,CAAC,qBAAqB,CAAC,CAkBhG"}
|