@aws/ml-container-creator 0.9.1 → 0.10.3
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/LICENSE-THIRD-PARTY +9304 -0
- package/bin/cli.js +2 -0
- package/config/bootstrap-e2e-stack.json +341 -0
- package/config/bootstrap-stack.json +40 -3
- package/config/parameter-schema-v2.json +2049 -0
- package/config/tune-catalog.json +1781 -0
- package/infra/ci-harness/buildspec.yml +1 -0
- package/infra/ci-harness/lambda/path-prover/brain.ts +306 -0
- package/infra/ci-harness/lambda/path-prover/write-results.ts +152 -0
- package/infra/ci-harness/lib/ci-harness-stack.ts +837 -7
- package/infra/ci-harness/state-machines/path-prover.asl.json +496 -0
- package/package.json +53 -68
- package/servers/base-image-picker/index.js +121 -121
- package/servers/e2e-status/index.js +297 -0
- package/servers/e2e-status/manifest.json +14 -0
- package/servers/e2e-status/package.json +15 -0
- package/servers/endpoint-picker/LICENSE +202 -0
- package/servers/endpoint-picker/index.js +536 -0
- package/servers/endpoint-picker/manifest.json +14 -0
- package/servers/endpoint-picker/package.json +18 -0
- package/servers/hyperpod-cluster-picker/index.js +125 -125
- package/servers/instance-sizer/index.js +138 -138
- package/servers/instance-sizer/lib/instance-ranker.js +76 -76
- package/servers/instance-sizer/lib/model-resolver.js +61 -61
- package/servers/instance-sizer/lib/quota-resolver.js +113 -113
- package/servers/instance-sizer/lib/vram-estimator.js +31 -31
- package/servers/lib/bedrock-client.js +38 -38
- package/servers/lib/catalogs/jumpstart-public.json +101 -16
- package/servers/lib/catalogs/model-servers.json +201 -3
- package/servers/lib/catalogs/models.json +182 -26
- package/servers/lib/custom-validators.js +13 -13
- package/servers/lib/dynamic-resolver.js +4 -4
- package/servers/marketplace-picker/index.js +342 -0
- package/servers/marketplace-picker/manifest.json +14 -0
- package/servers/marketplace-picker/package.json +18 -0
- package/servers/model-picker/index.js +382 -382
- package/servers/region-picker/index.js +56 -56
- package/servers/workload-picker/LICENSE +202 -0
- package/servers/workload-picker/catalogs/workload-profiles.json +67 -0
- package/servers/workload-picker/index.js +171 -0
- package/servers/workload-picker/manifest.json +16 -0
- package/servers/workload-picker/package.json +16 -0
- package/src/app.js +4 -390
- package/src/lib/bootstrap-command-handler.js +710 -1148
- package/src/lib/bootstrap-config.js +36 -0
- package/src/lib/bootstrap-profile-manager.js +641 -0
- package/src/lib/bootstrap-provisioners.js +421 -0
- package/src/lib/ci-register-helpers.js +74 -0
- package/src/lib/config-loader.js +408 -0
- package/src/lib/config-manager.js +66 -1685
- package/src/lib/config-mcp-client.js +118 -0
- package/src/lib/config-validator.js +634 -0
- package/src/lib/cuda-resolver.js +149 -0
- package/src/lib/e2e-catalog-validator.js +251 -3
- package/src/lib/e2e-ci-recorder.js +103 -0
- package/src/lib/generated/cli-options.js +315 -311
- package/src/lib/generated/parameter-matrix.js +671 -0
- package/src/lib/generated/validation-rules.js +71 -71
- package/src/lib/marketplace-flow.js +276 -0
- package/src/lib/mcp-query-runner.js +768 -0
- package/src/lib/parameter-schema-validator.js +62 -18
- package/src/lib/path-prover-brain.js +607 -0
- package/src/lib/prompt-runner.js +41 -1504
- package/src/lib/prompts/feature-prompts.js +172 -0
- package/src/lib/prompts/index.js +48 -0
- package/src/lib/prompts/infrastructure-prompts.js +690 -0
- package/src/lib/prompts/model-prompts.js +552 -0
- package/src/lib/prompts/project-prompts.js +82 -0
- package/src/lib/prompts.js +2 -1446
- package/src/lib/registry-command-handler.js +135 -3
- package/src/lib/secrets-prompt-runner.js +251 -0
- package/src/lib/template-variable-resolver.js +422 -0
- package/src/lib/tune-catalog-validator.js +37 -4
- package/templates/Dockerfile +9 -0
- package/templates/code/adapter_sidecar.py +444 -0
- package/templates/code/serve +6 -0
- package/templates/code/serve.d/vllm.ejs +1 -1
- package/templates/do/.benchmark_writer.py +1476 -0
- package/templates/do/.tune_helper.py +982 -57
- package/templates/do/__pycache__/.benchmark_writer.cpython-312.pyc +0 -0
- package/templates/do/adapter +149 -0
- package/templates/do/benchmark +639 -85
- package/templates/do/config +108 -5
- package/templates/do/deploy.d/managed-inference.ejs +192 -11
- package/templates/do/optimize +106 -37
- package/templates/do/register +89 -0
- package/templates/do/test +13 -0
- package/templates/do/tune +378 -59
- package/templates/do/validate +44 -4
- package/config/parameter-schema.json +0 -88
|
@@ -0,0 +1,496 @@
|
|
|
1
|
+
{
|
|
2
|
+
"Comment": "Path Prover Orchestrator — proves untested configuration paths end-to-end. Loops: Brain → stages → PickNext → Brain|End. Budget-controlled.",
|
|
3
|
+
"StartAt": "Brain",
|
|
4
|
+
"States": {
|
|
5
|
+
"Brain": {
|
|
6
|
+
"Type": "Task",
|
|
7
|
+
"Resource": "${BrainFunctionArn}",
|
|
8
|
+
"Parameters": {
|
|
9
|
+
"action": "getNextConfig",
|
|
10
|
+
"iteration.$": "$.iteration",
|
|
11
|
+
"budgetSpent.$": "$.budgetSpent",
|
|
12
|
+
"maxProvesPerRun.$": "$.maxProvesPerRun",
|
|
13
|
+
"maxCostPerRun.$": "$.maxCostPerRun",
|
|
14
|
+
"previousResults.$": "$.previousResults",
|
|
15
|
+
"priorityConfigPath.$": "$.priorityConfigPath"
|
|
16
|
+
},
|
|
17
|
+
"ResultPath": "$.brainResult",
|
|
18
|
+
"Retry": [
|
|
19
|
+
{
|
|
20
|
+
"ErrorEquals": ["Lambda.ServiceException", "Lambda.AWSLambdaException"],
|
|
21
|
+
"IntervalSeconds": 5,
|
|
22
|
+
"MaxAttempts": 2,
|
|
23
|
+
"BackoffRate": 2.0
|
|
24
|
+
}
|
|
25
|
+
],
|
|
26
|
+
"Catch": [
|
|
27
|
+
{
|
|
28
|
+
"ErrorEquals": ["States.ALL"],
|
|
29
|
+
"ResultPath": "$.error",
|
|
30
|
+
"Next": "ClassifyFailure"
|
|
31
|
+
}
|
|
32
|
+
],
|
|
33
|
+
"Next": "CheckBrainDone"
|
|
34
|
+
},
|
|
35
|
+
|
|
36
|
+
"CheckBrainDone": {
|
|
37
|
+
"Type": "Choice",
|
|
38
|
+
"Choices": [
|
|
39
|
+
{
|
|
40
|
+
"Variable": "$.brainResult.done",
|
|
41
|
+
"BooleanEquals": true,
|
|
42
|
+
"Next": "End"
|
|
43
|
+
}
|
|
44
|
+
],
|
|
45
|
+
"Default": "PrepareStageInput"
|
|
46
|
+
},
|
|
47
|
+
|
|
48
|
+
"PrepareStageInput": {
|
|
49
|
+
"Type": "Pass",
|
|
50
|
+
"Parameters": {
|
|
51
|
+
"iteration.$": "$.iteration",
|
|
52
|
+
"budgetSpent.$": "$.budgetSpent",
|
|
53
|
+
"maxProvesPerRun.$": "$.maxProvesPerRun",
|
|
54
|
+
"maxCostPerRun.$": "$.maxCostPerRun",
|
|
55
|
+
"previousResults.$": "$.previousResults",
|
|
56
|
+
"brainResult.$": "$.brainResult",
|
|
57
|
+
"currentConfig.$": "$.brainResult.next",
|
|
58
|
+
"tuneRequested.$": "$.brainResult.tuneRequested",
|
|
59
|
+
"priorityConfigPath.$": "$.priorityConfigPath"
|
|
60
|
+
},
|
|
61
|
+
"Next": "GenerateBuildPush"
|
|
62
|
+
},
|
|
63
|
+
|
|
64
|
+
"GenerateBuildPush": {
|
|
65
|
+
"Type": "Task",
|
|
66
|
+
"Resource": "arn:aws:states:::codebuild:startBuild",
|
|
67
|
+
"Parameters": {
|
|
68
|
+
"ProjectName": "${CodeBuildProjectName}",
|
|
69
|
+
"EnvironmentVariablesOverride": [
|
|
70
|
+
{
|
|
71
|
+
"Name": "CONFIG_JSON",
|
|
72
|
+
"Value.$": "States.JsonToString($.currentConfig)",
|
|
73
|
+
"Type": "PLAINTEXT"
|
|
74
|
+
},
|
|
75
|
+
{
|
|
76
|
+
"Name": "CI_STAGE",
|
|
77
|
+
"Value": "path-prove-generate-build-push",
|
|
78
|
+
"Type": "PLAINTEXT"
|
|
79
|
+
},
|
|
80
|
+
{
|
|
81
|
+
"Name": "PATH_PROVER_MODE",
|
|
82
|
+
"Value": "true",
|
|
83
|
+
"Type": "PLAINTEXT"
|
|
84
|
+
}
|
|
85
|
+
]
|
|
86
|
+
},
|
|
87
|
+
"ResultPath": "$.generateBuildPushResult",
|
|
88
|
+
"Retry": [
|
|
89
|
+
{
|
|
90
|
+
"ErrorEquals": ["States.TaskFailed"],
|
|
91
|
+
"IntervalSeconds": 30,
|
|
92
|
+
"MaxAttempts": 1,
|
|
93
|
+
"BackoffRate": 2.0
|
|
94
|
+
}
|
|
95
|
+
],
|
|
96
|
+
"Catch": [
|
|
97
|
+
{
|
|
98
|
+
"ErrorEquals": ["States.ALL"],
|
|
99
|
+
"ResultPath": "$.error",
|
|
100
|
+
"Next": "ClassifyFailure"
|
|
101
|
+
}
|
|
102
|
+
],
|
|
103
|
+
"Next": "DeployTest"
|
|
104
|
+
},
|
|
105
|
+
|
|
106
|
+
"DeployTest": {
|
|
107
|
+
"Type": "Task",
|
|
108
|
+
"Resource": "arn:aws:states:::codebuild:startBuild",
|
|
109
|
+
"Parameters": {
|
|
110
|
+
"ProjectName": "${CodeBuildProjectName}",
|
|
111
|
+
"EnvironmentVariablesOverride": [
|
|
112
|
+
{
|
|
113
|
+
"Name": "CONFIG_JSON",
|
|
114
|
+
"Value.$": "States.JsonToString($.currentConfig)",
|
|
115
|
+
"Type": "PLAINTEXT"
|
|
116
|
+
},
|
|
117
|
+
{
|
|
118
|
+
"Name": "CI_STAGE",
|
|
119
|
+
"Value": "path-prove-deploy-test",
|
|
120
|
+
"Type": "PLAINTEXT"
|
|
121
|
+
},
|
|
122
|
+
{
|
|
123
|
+
"Name": "PATH_PROVER_MODE",
|
|
124
|
+
"Value": "true",
|
|
125
|
+
"Type": "PLAINTEXT"
|
|
126
|
+
}
|
|
127
|
+
]
|
|
128
|
+
},
|
|
129
|
+
"ResultPath": "$.deployTestResult",
|
|
130
|
+
"Retry": [
|
|
131
|
+
{
|
|
132
|
+
"ErrorEquals": ["States.TaskFailed"],
|
|
133
|
+
"IntervalSeconds": 30,
|
|
134
|
+
"MaxAttempts": 1,
|
|
135
|
+
"BackoffRate": 2.0
|
|
136
|
+
}
|
|
137
|
+
],
|
|
138
|
+
"Catch": [
|
|
139
|
+
{
|
|
140
|
+
"ErrorEquals": ["States.ALL"],
|
|
141
|
+
"ResultPath": "$.error",
|
|
142
|
+
"Next": "ClassifyFailure"
|
|
143
|
+
}
|
|
144
|
+
],
|
|
145
|
+
"Next": "CheckTune"
|
|
146
|
+
},
|
|
147
|
+
|
|
148
|
+
"CheckTune": {
|
|
149
|
+
"Type": "Choice",
|
|
150
|
+
"Choices": [
|
|
151
|
+
{
|
|
152
|
+
"Variable": "$.tuneRequested",
|
|
153
|
+
"BooleanEquals": true,
|
|
154
|
+
"Next": "TuneAdapterTest"
|
|
155
|
+
}
|
|
156
|
+
],
|
|
157
|
+
"Default": "Benchmark"
|
|
158
|
+
},
|
|
159
|
+
|
|
160
|
+
"TuneAdapterTest": {
|
|
161
|
+
"Type": "Task",
|
|
162
|
+
"Resource": "arn:aws:states:::codebuild:startBuild",
|
|
163
|
+
"Parameters": {
|
|
164
|
+
"ProjectName": "${CodeBuildProjectName}",
|
|
165
|
+
"EnvironmentVariablesOverride": [
|
|
166
|
+
{
|
|
167
|
+
"Name": "CONFIG_JSON",
|
|
168
|
+
"Value.$": "States.JsonToString($.currentConfig)",
|
|
169
|
+
"Type": "PLAINTEXT"
|
|
170
|
+
},
|
|
171
|
+
{
|
|
172
|
+
"Name": "CI_STAGE",
|
|
173
|
+
"Value": "path-prove-tune-adapter-test",
|
|
174
|
+
"Type": "PLAINTEXT"
|
|
175
|
+
},
|
|
176
|
+
{
|
|
177
|
+
"Name": "PATH_PROVER_MODE",
|
|
178
|
+
"Value": "true",
|
|
179
|
+
"Type": "PLAINTEXT"
|
|
180
|
+
}
|
|
181
|
+
]
|
|
182
|
+
},
|
|
183
|
+
"ResultPath": "$.tuneAdapterTestResult",
|
|
184
|
+
"Retry": [
|
|
185
|
+
{
|
|
186
|
+
"ErrorEquals": ["States.TaskFailed"],
|
|
187
|
+
"IntervalSeconds": 30,
|
|
188
|
+
"MaxAttempts": 1,
|
|
189
|
+
"BackoffRate": 2.0
|
|
190
|
+
}
|
|
191
|
+
],
|
|
192
|
+
"Catch": [
|
|
193
|
+
{
|
|
194
|
+
"ErrorEquals": ["States.ALL"],
|
|
195
|
+
"ResultPath": "$.error",
|
|
196
|
+
"Next": "ClassifyFailure"
|
|
197
|
+
}
|
|
198
|
+
],
|
|
199
|
+
"Next": "Benchmark"
|
|
200
|
+
},
|
|
201
|
+
|
|
202
|
+
"Benchmark": {
|
|
203
|
+
"Type": "Task",
|
|
204
|
+
"Resource": "arn:aws:states:::codebuild:startBuild",
|
|
205
|
+
"Parameters": {
|
|
206
|
+
"ProjectName": "${CodeBuildProjectName}",
|
|
207
|
+
"EnvironmentVariablesOverride": [
|
|
208
|
+
{
|
|
209
|
+
"Name": "CONFIG_JSON",
|
|
210
|
+
"Value.$": "States.JsonToString($.currentConfig)",
|
|
211
|
+
"Type": "PLAINTEXT"
|
|
212
|
+
},
|
|
213
|
+
{
|
|
214
|
+
"Name": "CI_STAGE",
|
|
215
|
+
"Value": "path-prove-benchmark",
|
|
216
|
+
"Type": "PLAINTEXT"
|
|
217
|
+
},
|
|
218
|
+
{
|
|
219
|
+
"Name": "PATH_PROVER_MODE",
|
|
220
|
+
"Value": "true",
|
|
221
|
+
"Type": "PLAINTEXT"
|
|
222
|
+
}
|
|
223
|
+
]
|
|
224
|
+
},
|
|
225
|
+
"ResultPath": "$.benchmarkResult",
|
|
226
|
+
"Retry": [
|
|
227
|
+
{
|
|
228
|
+
"ErrorEquals": ["States.TaskFailed"],
|
|
229
|
+
"IntervalSeconds": 30,
|
|
230
|
+
"MaxAttempts": 1,
|
|
231
|
+
"BackoffRate": 2.0
|
|
232
|
+
}
|
|
233
|
+
],
|
|
234
|
+
"Catch": [
|
|
235
|
+
{
|
|
236
|
+
"ErrorEquals": ["States.ALL"],
|
|
237
|
+
"ResultPath": "$.error",
|
|
238
|
+
"Next": "ClassifyFailure"
|
|
239
|
+
}
|
|
240
|
+
],
|
|
241
|
+
"Next": "WriteResults"
|
|
242
|
+
},
|
|
243
|
+
|
|
244
|
+
"WriteResults": {
|
|
245
|
+
"Type": "Task",
|
|
246
|
+
"Resource": "${WriteResultsFunctionArn}",
|
|
247
|
+
"Parameters": {
|
|
248
|
+
"action": "writeResults",
|
|
249
|
+
"config.$": "$.currentConfig",
|
|
250
|
+
"benchmarkResult.$": "$.benchmarkResult",
|
|
251
|
+
"runType": "path_prove"
|
|
252
|
+
},
|
|
253
|
+
"ResultPath": "$.writeResult",
|
|
254
|
+
"Retry": [
|
|
255
|
+
{
|
|
256
|
+
"ErrorEquals": ["Lambda.ServiceException", "Lambda.AWSLambdaException"],
|
|
257
|
+
"IntervalSeconds": 5,
|
|
258
|
+
"MaxAttempts": 2,
|
|
259
|
+
"BackoffRate": 2.0
|
|
260
|
+
}
|
|
261
|
+
],
|
|
262
|
+
"Catch": [
|
|
263
|
+
{
|
|
264
|
+
"ErrorEquals": ["States.ALL"],
|
|
265
|
+
"ResultPath": "$.error",
|
|
266
|
+
"Next": "ClassifyFailure"
|
|
267
|
+
}
|
|
268
|
+
],
|
|
269
|
+
"Next": "Clean"
|
|
270
|
+
},
|
|
271
|
+
|
|
272
|
+
"Clean": {
|
|
273
|
+
"Type": "Task",
|
|
274
|
+
"Resource": "arn:aws:states:::codebuild:startBuild",
|
|
275
|
+
"Parameters": {
|
|
276
|
+
"ProjectName": "${CodeBuildProjectName}",
|
|
277
|
+
"EnvironmentVariablesOverride": [
|
|
278
|
+
{
|
|
279
|
+
"Name": "CONFIG_JSON",
|
|
280
|
+
"Value.$": "States.JsonToString($.currentConfig)",
|
|
281
|
+
"Type": "PLAINTEXT"
|
|
282
|
+
},
|
|
283
|
+
{
|
|
284
|
+
"Name": "CI_STAGE",
|
|
285
|
+
"Value": "path-prove-clean",
|
|
286
|
+
"Type": "PLAINTEXT"
|
|
287
|
+
},
|
|
288
|
+
{
|
|
289
|
+
"Name": "PATH_PROVER_MODE",
|
|
290
|
+
"Value": "true",
|
|
291
|
+
"Type": "PLAINTEXT"
|
|
292
|
+
}
|
|
293
|
+
]
|
|
294
|
+
},
|
|
295
|
+
"ResultPath": "$.cleanResult",
|
|
296
|
+
"Retry": [
|
|
297
|
+
{
|
|
298
|
+
"ErrorEquals": ["States.TaskFailed"],
|
|
299
|
+
"IntervalSeconds": 10,
|
|
300
|
+
"MaxAttempts": 2,
|
|
301
|
+
"BackoffRate": 2.0
|
|
302
|
+
}
|
|
303
|
+
],
|
|
304
|
+
"Catch": [
|
|
305
|
+
{
|
|
306
|
+
"ErrorEquals": ["States.ALL"],
|
|
307
|
+
"ResultPath": "$.cleanError",
|
|
308
|
+
"Next": "PickNext"
|
|
309
|
+
}
|
|
310
|
+
],
|
|
311
|
+
"Next": "PickNext"
|
|
312
|
+
},
|
|
313
|
+
|
|
314
|
+
"PickNext": {
|
|
315
|
+
"Type": "Task",
|
|
316
|
+
"Resource": "${BrainFunctionArn}",
|
|
317
|
+
"Parameters": {
|
|
318
|
+
"action": "pickNext",
|
|
319
|
+
"iteration.$": "$.iteration",
|
|
320
|
+
"budgetSpent.$": "$.budgetSpent",
|
|
321
|
+
"maxProvesPerRun.$": "$.maxProvesPerRun",
|
|
322
|
+
"maxCostPerRun.$": "$.maxCostPerRun",
|
|
323
|
+
"currentConfig.$": "$.currentConfig",
|
|
324
|
+
"previousResults.$": "$.previousResults",
|
|
325
|
+
"lastResult": "success",
|
|
326
|
+
"priorityConfigPath.$": "$.priorityConfigPath"
|
|
327
|
+
},
|
|
328
|
+
"ResultPath": "$.pickNextResult",
|
|
329
|
+
"Retry": [
|
|
330
|
+
{
|
|
331
|
+
"ErrorEquals": ["Lambda.ServiceException", "Lambda.AWSLambdaException"],
|
|
332
|
+
"IntervalSeconds": 5,
|
|
333
|
+
"MaxAttempts": 2,
|
|
334
|
+
"BackoffRate": 2.0
|
|
335
|
+
}
|
|
336
|
+
],
|
|
337
|
+
"Catch": [
|
|
338
|
+
{
|
|
339
|
+
"ErrorEquals": ["States.ALL"],
|
|
340
|
+
"ResultPath": "$.error",
|
|
341
|
+
"Next": "End"
|
|
342
|
+
}
|
|
343
|
+
],
|
|
344
|
+
"Next": "CheckDone"
|
|
345
|
+
},
|
|
346
|
+
|
|
347
|
+
"CheckDone": {
|
|
348
|
+
"Type": "Choice",
|
|
349
|
+
"Choices": [
|
|
350
|
+
{
|
|
351
|
+
"Variable": "$.pickNextResult.done",
|
|
352
|
+
"BooleanEquals": true,
|
|
353
|
+
"Next": "End"
|
|
354
|
+
}
|
|
355
|
+
],
|
|
356
|
+
"Default": "UpdateIterationState"
|
|
357
|
+
},
|
|
358
|
+
|
|
359
|
+
"UpdateIterationState": {
|
|
360
|
+
"Type": "Pass",
|
|
361
|
+
"Parameters": {
|
|
362
|
+
"iteration.$": "$.pickNextResult.iteration",
|
|
363
|
+
"budgetSpent.$": "$.pickNextResult.budgetSpent",
|
|
364
|
+
"maxProvesPerRun.$": "$.maxProvesPerRun",
|
|
365
|
+
"maxCostPerRun.$": "$.maxCostPerRun",
|
|
366
|
+
"previousResults.$": "$.pickNextResult.previousResults",
|
|
367
|
+
"brainResult.$": "$.pickNextResult",
|
|
368
|
+
"priorityConfigPath.$": "$.priorityConfigPath"
|
|
369
|
+
},
|
|
370
|
+
"Next": "CheckBrainDone"
|
|
371
|
+
},
|
|
372
|
+
|
|
373
|
+
"ClassifyFailure": {
|
|
374
|
+
"Type": "Task",
|
|
375
|
+
"Resource": "${ClassifyFailureFunctionArn}",
|
|
376
|
+
"Parameters": {
|
|
377
|
+
"action": "classifyFailure",
|
|
378
|
+
"error.$": "$.error",
|
|
379
|
+
"config.$": "$.currentConfig",
|
|
380
|
+
"iteration.$": "$.iteration"
|
|
381
|
+
},
|
|
382
|
+
"ResultPath": "$.classification",
|
|
383
|
+
"Retry": [
|
|
384
|
+
{
|
|
385
|
+
"ErrorEquals": ["Lambda.ServiceException", "Lambda.AWSLambdaException"],
|
|
386
|
+
"IntervalSeconds": 5,
|
|
387
|
+
"MaxAttempts": 2,
|
|
388
|
+
"BackoffRate": 2.0
|
|
389
|
+
}
|
|
390
|
+
],
|
|
391
|
+
"Next": "WriteFailureRecord"
|
|
392
|
+
},
|
|
393
|
+
|
|
394
|
+
"WriteFailureRecord": {
|
|
395
|
+
"Type": "Task",
|
|
396
|
+
"Resource": "${WriteResultsFunctionArn}",
|
|
397
|
+
"Parameters": {
|
|
398
|
+
"action": "writeFailure",
|
|
399
|
+
"config.$": "$.currentConfig",
|
|
400
|
+
"error.$": "$.error",
|
|
401
|
+
"classification.$": "$.classification",
|
|
402
|
+
"runType": "path_prove"
|
|
403
|
+
},
|
|
404
|
+
"ResultPath": "$.failureWriteResult",
|
|
405
|
+
"Retry": [
|
|
406
|
+
{
|
|
407
|
+
"ErrorEquals": ["Lambda.ServiceException", "Lambda.AWSLambdaException"],
|
|
408
|
+
"IntervalSeconds": 5,
|
|
409
|
+
"MaxAttempts": 2,
|
|
410
|
+
"BackoffRate": 2.0
|
|
411
|
+
}
|
|
412
|
+
],
|
|
413
|
+
"Next": "CleanAfterFailure"
|
|
414
|
+
},
|
|
415
|
+
|
|
416
|
+
"CleanAfterFailure": {
|
|
417
|
+
"Type": "Task",
|
|
418
|
+
"Resource": "arn:aws:states:::codebuild:startBuild",
|
|
419
|
+
"Parameters": {
|
|
420
|
+
"ProjectName": "${CodeBuildProjectName}",
|
|
421
|
+
"EnvironmentVariablesOverride": [
|
|
422
|
+
{
|
|
423
|
+
"Name": "CONFIG_JSON",
|
|
424
|
+
"Value.$": "States.JsonToString($.currentConfig)",
|
|
425
|
+
"Type": "PLAINTEXT"
|
|
426
|
+
},
|
|
427
|
+
{
|
|
428
|
+
"Name": "CI_STAGE",
|
|
429
|
+
"Value": "path-prove-clean",
|
|
430
|
+
"Type": "PLAINTEXT"
|
|
431
|
+
},
|
|
432
|
+
{
|
|
433
|
+
"Name": "PATH_PROVER_MODE",
|
|
434
|
+
"Value": "true",
|
|
435
|
+
"Type": "PLAINTEXT"
|
|
436
|
+
}
|
|
437
|
+
]
|
|
438
|
+
},
|
|
439
|
+
"ResultPath": "$.failureCleanResult",
|
|
440
|
+
"Retry": [
|
|
441
|
+
{
|
|
442
|
+
"ErrorEquals": ["States.TaskFailed"],
|
|
443
|
+
"IntervalSeconds": 10,
|
|
444
|
+
"MaxAttempts": 2,
|
|
445
|
+
"BackoffRate": 2.0
|
|
446
|
+
}
|
|
447
|
+
],
|
|
448
|
+
"Catch": [
|
|
449
|
+
{
|
|
450
|
+
"ErrorEquals": ["States.ALL"],
|
|
451
|
+
"ResultPath": "$.failureCleanError",
|
|
452
|
+
"Next": "PickNextAfterFailure"
|
|
453
|
+
}
|
|
454
|
+
],
|
|
455
|
+
"Next": "PickNextAfterFailure"
|
|
456
|
+
},
|
|
457
|
+
|
|
458
|
+
"PickNextAfterFailure": {
|
|
459
|
+
"Type": "Task",
|
|
460
|
+
"Resource": "${BrainFunctionArn}",
|
|
461
|
+
"Parameters": {
|
|
462
|
+
"action": "pickNext",
|
|
463
|
+
"iteration.$": "$.iteration",
|
|
464
|
+
"budgetSpent.$": "$.budgetSpent",
|
|
465
|
+
"maxProvesPerRun.$": "$.maxProvesPerRun",
|
|
466
|
+
"maxCostPerRun.$": "$.maxCostPerRun",
|
|
467
|
+
"currentConfig.$": "$.currentConfig",
|
|
468
|
+
"previousResults.$": "$.previousResults",
|
|
469
|
+
"lastResult": "failure",
|
|
470
|
+
"classification.$": "$.classification",
|
|
471
|
+
"priorityConfigPath.$": "$.priorityConfigPath"
|
|
472
|
+
},
|
|
473
|
+
"ResultPath": "$.pickNextResult",
|
|
474
|
+
"Retry": [
|
|
475
|
+
{
|
|
476
|
+
"ErrorEquals": ["Lambda.ServiceException", "Lambda.AWSLambdaException"],
|
|
477
|
+
"IntervalSeconds": 5,
|
|
478
|
+
"MaxAttempts": 2,
|
|
479
|
+
"BackoffRate": 2.0
|
|
480
|
+
}
|
|
481
|
+
],
|
|
482
|
+
"Catch": [
|
|
483
|
+
{
|
|
484
|
+
"ErrorEquals": ["States.ALL"],
|
|
485
|
+
"ResultPath": "$.error",
|
|
486
|
+
"Next": "End"
|
|
487
|
+
}
|
|
488
|
+
],
|
|
489
|
+
"Next": "CheckDone"
|
|
490
|
+
},
|
|
491
|
+
|
|
492
|
+
"End": {
|
|
493
|
+
"Type": "Succeed"
|
|
494
|
+
}
|
|
495
|
+
}
|
|
496
|
+
}
|
package/package.json
CHANGED
|
@@ -1,16 +1,18 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@aws/ml-container-creator",
|
|
3
|
-
"version": "0.
|
|
4
|
-
"description": "
|
|
5
|
-
"
|
|
6
|
-
"main": "src/app.js",
|
|
3
|
+
"version": "0.10.3",
|
|
4
|
+
"description": "Build and deploy custom ML containers on AWS SageMaker with minimal configuration.",
|
|
5
|
+
"main": "src/index.js",
|
|
7
6
|
"bin": {
|
|
8
|
-
"ml-container-creator": "./bin/cli.js"
|
|
9
|
-
|
|
10
|
-
"publishConfig": {
|
|
11
|
-
"access": "public"
|
|
7
|
+
"ml-container-creator": "./bin/cli.js",
|
|
8
|
+
"mcc": "./bin/cli.js"
|
|
12
9
|
},
|
|
13
|
-
"
|
|
10
|
+
"keywords": [
|
|
11
|
+
"sagemaker",
|
|
12
|
+
"docker",
|
|
13
|
+
"ml",
|
|
14
|
+
"containers"
|
|
15
|
+
],
|
|
14
16
|
"repository": {
|
|
15
17
|
"type": "git",
|
|
16
18
|
"url": "git+https://github.com/awslabs/ml-container-creator.git"
|
|
@@ -36,6 +38,19 @@
|
|
|
36
38
|
"servers/instance-sizer/index.js",
|
|
37
39
|
"servers/instance-sizer/manifest.json",
|
|
38
40
|
"servers/instance-sizer/package.json",
|
|
41
|
+
"servers/endpoint-picker/index.js",
|
|
42
|
+
"servers/endpoint-picker/manifest.json",
|
|
43
|
+
"servers/endpoint-picker/package.json",
|
|
44
|
+
"servers/marketplace-picker/index.js",
|
|
45
|
+
"servers/marketplace-picker/manifest.json",
|
|
46
|
+
"servers/marketplace-picker/package.json",
|
|
47
|
+
"servers/e2e-status/index.js",
|
|
48
|
+
"servers/e2e-status/manifest.json",
|
|
49
|
+
"servers/e2e-status/package.json",
|
|
50
|
+
"servers/workload-picker/catalogs/",
|
|
51
|
+
"servers/workload-picker/index.js",
|
|
52
|
+
"servers/workload-picker/manifest.json",
|
|
53
|
+
"servers/workload-picker/package.json",
|
|
39
54
|
"servers/lib/bedrock-client.js",
|
|
40
55
|
"servers/lib/custom-validators.js",
|
|
41
56
|
"servers/lib/dynamic-resolver.js",
|
|
@@ -44,65 +59,49 @@
|
|
|
44
59
|
"servers/README.md",
|
|
45
60
|
"config/defaults.json",
|
|
46
61
|
"config/bootstrap-stack.json",
|
|
47
|
-
"config/
|
|
62
|
+
"config/bootstrap-e2e-stack.json",
|
|
63
|
+
"config/parameter-schema-v2.json",
|
|
64
|
+
"config/tune-catalog.json",
|
|
48
65
|
"config/presets/",
|
|
49
66
|
"infra/ci-harness/bin/",
|
|
50
67
|
"infra/ci-harness/lambda/",
|
|
51
68
|
"infra/ci-harness/lib/",
|
|
69
|
+
"infra/ci-harness/state-machines/",
|
|
52
70
|
"infra/ci-harness/buildspec.yml",
|
|
53
71
|
"infra/ci-harness/cdk.json",
|
|
54
72
|
"infra/ci-harness/package.json",
|
|
55
73
|
"infra/ci-harness/package-lock.json",
|
|
56
74
|
"infra/ci-harness/tsconfig.json",
|
|
75
|
+
"infra/ci-harness/state-machines/",
|
|
57
76
|
"README.md",
|
|
58
77
|
"LICENSE",
|
|
59
78
|
"LICENSE-THIRD-PARTY",
|
|
60
79
|
"NOTICE"
|
|
61
80
|
],
|
|
62
|
-
"
|
|
63
|
-
|
|
64
|
-
|
|
65
|
-
"
|
|
66
|
-
"
|
|
67
|
-
|
|
68
|
-
"machine-learning",
|
|
69
|
-
"docker"
|
|
70
|
-
],
|
|
81
|
+
"type": "module",
|
|
82
|
+
"license": "Apache-2.0",
|
|
83
|
+
"engines": {
|
|
84
|
+
"node": ">=24.11.1",
|
|
85
|
+
"npm": ">=11.6.2"
|
|
86
|
+
},
|
|
71
87
|
"scripts": {
|
|
72
|
-
"test": "
|
|
73
|
-
"test:
|
|
74
|
-
"test:
|
|
75
|
-
"test:
|
|
76
|
-
"test:
|
|
77
|
-
"test:
|
|
78
|
-
"test:
|
|
79
|
-
"test:
|
|
80
|
-
"test:
|
|
81
|
-
"test:
|
|
82
|
-
"test:
|
|
83
|
-
"
|
|
84
|
-
"
|
|
85
|
-
"
|
|
86
|
-
"
|
|
87
|
-
"
|
|
88
|
-
"
|
|
89
|
-
"start": "ml-container-creator",
|
|
90
|
-
"dev": "npm link && ml-container-creator",
|
|
91
|
-
"clean": "rm -rf test-output-*",
|
|
92
|
-
"validate": "npm run lint && npm run test:all",
|
|
93
|
-
"validate:catalogs": "node scripts/validate-catalog-enums.js",
|
|
94
|
-
"validate:namespaces": "node scripts/validate-namespaces.js",
|
|
95
|
-
"docs:serve": "mkdocs serve",
|
|
96
|
-
"docs:build": "mkdocs build",
|
|
97
|
-
"docs:deploy": "mkdocs gh-deploy",
|
|
98
|
-
"docs:sync": "node scripts/sync-command-generator.js",
|
|
99
|
-
"docs:check": "node scripts/sync-command-generator.js --check",
|
|
100
|
-
"codegen": "node scripts/codegen-cli.js && node scripts/codegen-validator.js && node scripts/codegen-widget.js",
|
|
101
|
-
"codegen:check": "node scripts/codegen-cli.js && node scripts/codegen-validator.js && node scripts/codegen-widget.js && node scripts/codegen-parity.js",
|
|
102
|
-
"_sbom": "npm sbom --sbom-format spdx > sbom.json && for dir in servers/*/; do [ -f \"$dir/package.json\" ] && (cd \"$dir\" && npm sbom --sbom-format spdx > sbom.json); done",
|
|
103
|
-
"_licenses:review": "license-checker --production --exclude MIT,Apache-2.0,BSD-2-Clause,BSD-3-Clause,ISC,0BSD && for dir in servers/*/; do [ -f \"$dir/package.json\" ] && echo \"\\nChecking $dir\" && (cd \"$dir\" && npx license-checker --production --exclude MIT,Apache-2.0,BSD-2-Clause,BSD-3-Clause,ISC,0BSD); done",
|
|
104
|
-
"_licenses:csv": "license-checker --csv --out ./licenses.csv && for dir in servers/*/; do [ -f \"$dir/package.json\" ] && (cd \"$dir\" && npx license-checker --csv --out licenses.csv); done && cat servers/*/licenses.csv >> licenses.csv",
|
|
105
|
-
"_licenses:attribution": "generate-attribution && mv oss-attribution/attribution.txt LICENSE-THIRD-PARTY && rm -rf oss-attribution && for dir in servers/*/; do [ -f \"$dir/package.json\" ] && (cd \"$dir\" && generate-attribution && cat oss-attribution/attribution.txt >> ../../LICENSE-THIRD-PARTY && rm -rf oss-attribution); done"
|
|
88
|
+
"test": "mocha 'test/**/*.test.js' --recursive --timeout 30000",
|
|
89
|
+
"test:property": "mocha 'test/property/**/*.test.js' --recursive --timeout 60000",
|
|
90
|
+
"test:all": "npm run test && npm run test:property",
|
|
91
|
+
"test:fast": "mocha 'test/**/*.test.js' --recursive --timeout 15000 --parallel",
|
|
92
|
+
"test:unit": "mocha 'test/unit/**/*.test.js' --recursive --timeout 15000",
|
|
93
|
+
"test:integration": "mocha 'test/integration/**/*.test.js' --recursive --timeout 30000",
|
|
94
|
+
"test:servers": "node servers/region-picker/test.js && node servers/instance-sizer/test.js && node servers/workload-picker/test.js",
|
|
95
|
+
"test:watch": "mocha 'test/**/*.test.js' --recursive --timeout 30000 --watch",
|
|
96
|
+
"test:coverage": "nyc npm test",
|
|
97
|
+
"test:ci": "npm run lint && npm run test:all",
|
|
98
|
+
"test:perf": "node scripts/analyze-test-performance.js",
|
|
99
|
+
"lint": "eslint src/ servers/ bin/ --ext .js,.cjs,.mjs",
|
|
100
|
+
"lint:fix": "eslint src/ servers/ bin/ --ext .js,.cjs,.mjs --fix",
|
|
101
|
+
"codegen": "node scripts/codegen-cli.js && node scripts/codegen-validator.js && node scripts/codegen-widget.js && node scripts/codegen-parameter-matrix.js",
|
|
102
|
+
"validate:doc-commands": "node scripts/validate-docs-commands.js",
|
|
103
|
+
"sbom": "sbom --format spdx --output sbom.json",
|
|
104
|
+
"prepublishOnly": "npm run lint && npm run test:all"
|
|
106
105
|
},
|
|
107
106
|
"dependencies": {
|
|
108
107
|
"@inquirer/prompts": "^8.4.2",
|
|
@@ -126,19 +125,5 @@
|
|
|
126
125
|
"npm-force-resolutions": "^0.0.10",
|
|
127
126
|
"nyc": "^15.1.0",
|
|
128
127
|
"sbom": "^0.0.0"
|
|
129
|
-
}
|
|
130
|
-
"engines": {
|
|
131
|
-
"node": ">=24.11.1",
|
|
132
|
-
"npm": ">=11.6.2"
|
|
133
|
-
},
|
|
134
|
-
"overrides": {
|
|
135
|
-
"cross-spawn": "^7.0.3",
|
|
136
|
-
"diff": "^5.2.2",
|
|
137
|
-
"semver": "^7.5.4",
|
|
138
|
-
"got": "^11.8.5",
|
|
139
|
-
"qs": "^6.14.1",
|
|
140
|
-
"serialize-javascript": "^7.0.5"
|
|
141
|
-
},
|
|
142
|
-
"author": "Dan Ferguson",
|
|
143
|
-
"license": "Apache-2.0"
|
|
128
|
+
}
|
|
144
129
|
}
|