aws-cdk-neuronx-patterns 0.1.1 → 0.2.0
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/.jsii +237 -67
- package/AGENT.md +11 -0
- package/API.md +130 -0
- package/lib/base/aws-batch/neuronx-batch-compute-environment.js +1 -1
- package/lib/base/aws-batch/neuronx-batch-ecs-job-definition.js +1 -1
- package/lib/base/aws-batch/neuronx-batch.js +1 -1
- package/lib/base/aws-ecs-patterns/application-load-balanced-neuronx-service.js +2 -2
- package/lib/base/neuronx/deep-learning-containers.d.ts +21 -3
- package/lib/base/neuronx/deep-learning-containers.js +25 -7
- package/lib/base/neuronx/deep-learning-containers.test.js +80 -3
- package/lib/base/neuronx/model.js +2 -2
- package/lib/base/neuronx/neuron-optimized-machine-image.js +1 -1
- package/lib/base/neuronx/neuronx-instance-type.js +3 -3
- package/lib/base/neuronx-compiler/neuronx-compiler.js +1 -1
- package/lib/base/server-engine/vllm-engine/vllm-engine-argments.js +1 -1
- package/lib/vllm-nxd-inference/vllm-nxd-inference-compiler.js +5 -3
- package/lib/vllm-nxd-inference/vllm-nxd-inference-ecs-patterns.js +4 -4
- package/package.json +1 -1
package/.jsii
CHANGED
|
@@ -9518,7 +9518,7 @@
|
|
|
9518
9518
|
"kind": "interface",
|
|
9519
9519
|
"locationInModule": {
|
|
9520
9520
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
9521
|
-
"line":
|
|
9521
|
+
"line": 233
|
|
9522
9522
|
},
|
|
9523
9523
|
"name": "IVllmInferenceNeuronxImage",
|
|
9524
9524
|
"properties": [
|
|
@@ -9530,7 +9530,7 @@
|
|
|
9530
9530
|
"immutable": true,
|
|
9531
9531
|
"locationInModule": {
|
|
9532
9532
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
9533
|
-
"line":
|
|
9533
|
+
"line": 234
|
|
9534
9534
|
},
|
|
9535
9535
|
"name": "vllmVersion",
|
|
9536
9536
|
"type": {
|
|
@@ -11983,7 +11983,7 @@
|
|
|
11983
11983
|
"kind": "class",
|
|
11984
11984
|
"locationInModule": {
|
|
11985
11985
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
11986
|
-
"line":
|
|
11986
|
+
"line": 159
|
|
11987
11987
|
},
|
|
11988
11988
|
"methods": [
|
|
11989
11989
|
{
|
|
@@ -11992,7 +11992,7 @@
|
|
|
11992
11992
|
},
|
|
11993
11993
|
"locationInModule": {
|
|
11994
11994
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
11995
|
-
"line":
|
|
11995
|
+
"line": 144
|
|
11996
11996
|
},
|
|
11997
11997
|
"name": "fromNeuronSdkVersion",
|
|
11998
11998
|
"parameters": [
|
|
@@ -12039,7 +12039,7 @@
|
|
|
12039
12039
|
"immutable": true,
|
|
12040
12040
|
"locationInModule": {
|
|
12041
12041
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12042
|
-
"line":
|
|
12042
|
+
"line": 160
|
|
12043
12043
|
},
|
|
12044
12044
|
"name": "imageName",
|
|
12045
12045
|
"static": true,
|
|
@@ -12056,7 +12056,7 @@
|
|
|
12056
12056
|
"immutable": true,
|
|
12057
12057
|
"locationInModule": {
|
|
12058
12058
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12059
|
-
"line":
|
|
12059
|
+
"line": 226
|
|
12060
12060
|
},
|
|
12061
12061
|
"name": "LATEST",
|
|
12062
12062
|
"static": true,
|
|
@@ -12073,7 +12073,7 @@
|
|
|
12073
12073
|
"immutable": true,
|
|
12074
12074
|
"locationInModule": {
|
|
12075
12075
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12076
|
-
"line":
|
|
12076
|
+
"line": 135
|
|
12077
12077
|
},
|
|
12078
12078
|
"name": "SDK_2_18_0",
|
|
12079
12079
|
"static": true,
|
|
@@ -12090,7 +12090,7 @@
|
|
|
12090
12090
|
"immutable": true,
|
|
12091
12091
|
"locationInModule": {
|
|
12092
12092
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12093
|
-
"line":
|
|
12093
|
+
"line": 128
|
|
12094
12094
|
},
|
|
12095
12095
|
"name": "SDK_2_18_1",
|
|
12096
12096
|
"static": true,
|
|
@@ -12107,7 +12107,7 @@
|
|
|
12107
12107
|
"immutable": true,
|
|
12108
12108
|
"locationInModule": {
|
|
12109
12109
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12110
|
-
"line":
|
|
12110
|
+
"line": 121
|
|
12111
12111
|
},
|
|
12112
12112
|
"name": "SDK_2_18_2",
|
|
12113
12113
|
"static": true,
|
|
@@ -12124,7 +12124,7 @@
|
|
|
12124
12124
|
"immutable": true,
|
|
12125
12125
|
"locationInModule": {
|
|
12126
12126
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12127
|
-
"line":
|
|
12127
|
+
"line": 114
|
|
12128
12128
|
},
|
|
12129
12129
|
"name": "SDK_2_19_0",
|
|
12130
12130
|
"static": true,
|
|
@@ -12141,7 +12141,7 @@
|
|
|
12141
12141
|
"immutable": true,
|
|
12142
12142
|
"locationInModule": {
|
|
12143
12143
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12144
|
-
"line":
|
|
12144
|
+
"line": 107
|
|
12145
12145
|
},
|
|
12146
12146
|
"name": "SDK_2_19_1",
|
|
12147
12147
|
"static": true,
|
|
@@ -12158,7 +12158,7 @@
|
|
|
12158
12158
|
"immutable": true,
|
|
12159
12159
|
"locationInModule": {
|
|
12160
12160
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12161
|
-
"line":
|
|
12161
|
+
"line": 100
|
|
12162
12162
|
},
|
|
12163
12163
|
"name": "SDK_2_20_0",
|
|
12164
12164
|
"static": true,
|
|
@@ -12175,7 +12175,7 @@
|
|
|
12175
12175
|
"immutable": true,
|
|
12176
12176
|
"locationInModule": {
|
|
12177
12177
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12178
|
-
"line":
|
|
12178
|
+
"line": 93
|
|
12179
12179
|
},
|
|
12180
12180
|
"name": "SDK_2_20_1",
|
|
12181
12181
|
"static": true,
|
|
@@ -12192,7 +12192,7 @@
|
|
|
12192
12192
|
"immutable": true,
|
|
12193
12193
|
"locationInModule": {
|
|
12194
12194
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12195
|
-
"line":
|
|
12195
|
+
"line": 86
|
|
12196
12196
|
},
|
|
12197
12197
|
"name": "SDK_2_20_2",
|
|
12198
12198
|
"static": true,
|
|
@@ -12209,7 +12209,7 @@
|
|
|
12209
12209
|
"immutable": true,
|
|
12210
12210
|
"locationInModule": {
|
|
12211
12211
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12212
|
-
"line":
|
|
12212
|
+
"line": 218
|
|
12213
12213
|
},
|
|
12214
12214
|
"name": "SDK_2_20_2_PYTORCH_1_13_1",
|
|
12215
12215
|
"static": true,
|
|
@@ -12226,7 +12226,7 @@
|
|
|
12226
12226
|
"immutable": true,
|
|
12227
12227
|
"locationInModule": {
|
|
12228
12228
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12229
|
-
"line":
|
|
12229
|
+
"line": 210
|
|
12230
12230
|
},
|
|
12231
12231
|
"name": "SDK_2_20_2_PYTORCH_2_1_2",
|
|
12232
12232
|
"static": true,
|
|
@@ -12243,7 +12243,7 @@
|
|
|
12243
12243
|
"immutable": true,
|
|
12244
12244
|
"locationInModule": {
|
|
12245
12245
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12246
|
-
"line":
|
|
12246
|
+
"line": 79
|
|
12247
12247
|
},
|
|
12248
12248
|
"name": "SDK_2_21_0",
|
|
12249
12249
|
"static": true,
|
|
@@ -12260,7 +12260,7 @@
|
|
|
12260
12260
|
"immutable": true,
|
|
12261
12261
|
"locationInModule": {
|
|
12262
12262
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12263
|
-
"line":
|
|
12263
|
+
"line": 72
|
|
12264
12264
|
},
|
|
12265
12265
|
"name": "SDK_2_21_1",
|
|
12266
12266
|
"static": true,
|
|
@@ -12277,7 +12277,7 @@
|
|
|
12277
12277
|
"immutable": true,
|
|
12278
12278
|
"locationInModule": {
|
|
12279
12279
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12280
|
-
"line":
|
|
12280
|
+
"line": 202
|
|
12281
12281
|
},
|
|
12282
12282
|
"name": "SDK_2_22_0",
|
|
12283
12283
|
"static": true,
|
|
@@ -12294,7 +12294,7 @@
|
|
|
12294
12294
|
"immutable": true,
|
|
12295
12295
|
"locationInModule": {
|
|
12296
12296
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12297
|
-
"line":
|
|
12297
|
+
"line": 194
|
|
12298
12298
|
},
|
|
12299
12299
|
"name": "SDK_2_23_0",
|
|
12300
12300
|
"static": true,
|
|
@@ -12311,7 +12311,7 @@
|
|
|
12311
12311
|
"immutable": true,
|
|
12312
12312
|
"locationInModule": {
|
|
12313
12313
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12314
|
-
"line":
|
|
12314
|
+
"line": 51
|
|
12315
12315
|
},
|
|
12316
12316
|
"name": "SDK_2_24_0",
|
|
12317
12317
|
"static": true,
|
|
@@ -12328,7 +12328,7 @@
|
|
|
12328
12328
|
"immutable": true,
|
|
12329
12329
|
"locationInModule": {
|
|
12330
12330
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12331
|
-
"line":
|
|
12331
|
+
"line": 186
|
|
12332
12332
|
},
|
|
12333
12333
|
"name": "SDK_2_25_0",
|
|
12334
12334
|
"static": true,
|
|
@@ -12345,7 +12345,7 @@
|
|
|
12345
12345
|
"immutable": true,
|
|
12346
12346
|
"locationInModule": {
|
|
12347
12347
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12348
|
-
"line":
|
|
12348
|
+
"line": 178
|
|
12349
12349
|
},
|
|
12350
12350
|
"name": "SDK_2_26_1",
|
|
12351
12351
|
"static": true,
|
|
@@ -12362,7 +12362,7 @@
|
|
|
12362
12362
|
"immutable": true,
|
|
12363
12363
|
"locationInModule": {
|
|
12364
12364
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12365
|
-
"line":
|
|
12365
|
+
"line": 170
|
|
12366
12366
|
},
|
|
12367
12367
|
"name": "SDK_2_27_1",
|
|
12368
12368
|
"static": true,
|
|
@@ -12379,7 +12379,7 @@
|
|
|
12379
12379
|
"immutable": true,
|
|
12380
12380
|
"locationInModule": {
|
|
12381
12381
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12382
|
-
"line":
|
|
12382
|
+
"line": 162
|
|
12383
12383
|
},
|
|
12384
12384
|
"name": "SDK_2_28_0",
|
|
12385
12385
|
"static": true,
|
|
@@ -12421,7 +12421,7 @@
|
|
|
12421
12421
|
"kind": "class",
|
|
12422
12422
|
"locationInModule": {
|
|
12423
12423
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12424
|
-
"line":
|
|
12424
|
+
"line": 229
|
|
12425
12425
|
},
|
|
12426
12426
|
"methods": [
|
|
12427
12427
|
{
|
|
@@ -12430,7 +12430,7 @@
|
|
|
12430
12430
|
},
|
|
12431
12431
|
"locationInModule": {
|
|
12432
12432
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12433
|
-
"line":
|
|
12433
|
+
"line": 144
|
|
12434
12434
|
},
|
|
12435
12435
|
"name": "fromNeuronSdkVersion",
|
|
12436
12436
|
"parameters": [
|
|
@@ -12477,7 +12477,7 @@
|
|
|
12477
12477
|
"immutable": true,
|
|
12478
12478
|
"locationInModule": {
|
|
12479
12479
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12480
|
-
"line":
|
|
12480
|
+
"line": 230
|
|
12481
12481
|
},
|
|
12482
12482
|
"name": "imageName",
|
|
12483
12483
|
"static": true,
|
|
@@ -12494,7 +12494,7 @@
|
|
|
12494
12494
|
"immutable": true,
|
|
12495
12495
|
"locationInModule": {
|
|
12496
12496
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12497
|
-
"line":
|
|
12497
|
+
"line": 142
|
|
12498
12498
|
},
|
|
12499
12499
|
"name": "LATEST",
|
|
12500
12500
|
"static": true,
|
|
@@ -12511,7 +12511,7 @@
|
|
|
12511
12511
|
"immutable": true,
|
|
12512
12512
|
"locationInModule": {
|
|
12513
12513
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12514
|
-
"line":
|
|
12514
|
+
"line": 135
|
|
12515
12515
|
},
|
|
12516
12516
|
"name": "SDK_2_18_0",
|
|
12517
12517
|
"static": true,
|
|
@@ -12528,7 +12528,7 @@
|
|
|
12528
12528
|
"immutable": true,
|
|
12529
12529
|
"locationInModule": {
|
|
12530
12530
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12531
|
-
"line":
|
|
12531
|
+
"line": 128
|
|
12532
12532
|
},
|
|
12533
12533
|
"name": "SDK_2_18_1",
|
|
12534
12534
|
"static": true,
|
|
@@ -12545,7 +12545,7 @@
|
|
|
12545
12545
|
"immutable": true,
|
|
12546
12546
|
"locationInModule": {
|
|
12547
12547
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12548
|
-
"line":
|
|
12548
|
+
"line": 121
|
|
12549
12549
|
},
|
|
12550
12550
|
"name": "SDK_2_18_2",
|
|
12551
12551
|
"static": true,
|
|
@@ -12562,7 +12562,7 @@
|
|
|
12562
12562
|
"immutable": true,
|
|
12563
12563
|
"locationInModule": {
|
|
12564
12564
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12565
|
-
"line":
|
|
12565
|
+
"line": 114
|
|
12566
12566
|
},
|
|
12567
12567
|
"name": "SDK_2_19_0",
|
|
12568
12568
|
"static": true,
|
|
@@ -12579,7 +12579,7 @@
|
|
|
12579
12579
|
"immutable": true,
|
|
12580
12580
|
"locationInModule": {
|
|
12581
12581
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12582
|
-
"line":
|
|
12582
|
+
"line": 107
|
|
12583
12583
|
},
|
|
12584
12584
|
"name": "SDK_2_19_1",
|
|
12585
12585
|
"static": true,
|
|
@@ -12596,7 +12596,7 @@
|
|
|
12596
12596
|
"immutable": true,
|
|
12597
12597
|
"locationInModule": {
|
|
12598
12598
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12599
|
-
"line":
|
|
12599
|
+
"line": 100
|
|
12600
12600
|
},
|
|
12601
12601
|
"name": "SDK_2_20_0",
|
|
12602
12602
|
"static": true,
|
|
@@ -12613,7 +12613,7 @@
|
|
|
12613
12613
|
"immutable": true,
|
|
12614
12614
|
"locationInModule": {
|
|
12615
12615
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12616
|
-
"line":
|
|
12616
|
+
"line": 93
|
|
12617
12617
|
},
|
|
12618
12618
|
"name": "SDK_2_20_1",
|
|
12619
12619
|
"static": true,
|
|
@@ -12630,7 +12630,7 @@
|
|
|
12630
12630
|
"immutable": true,
|
|
12631
12631
|
"locationInModule": {
|
|
12632
12632
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12633
|
-
"line":
|
|
12633
|
+
"line": 86
|
|
12634
12634
|
},
|
|
12635
12635
|
"name": "SDK_2_20_2",
|
|
12636
12636
|
"static": true,
|
|
@@ -12647,7 +12647,7 @@
|
|
|
12647
12647
|
"immutable": true,
|
|
12648
12648
|
"locationInModule": {
|
|
12649
12649
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12650
|
-
"line":
|
|
12650
|
+
"line": 79
|
|
12651
12651
|
},
|
|
12652
12652
|
"name": "SDK_2_21_0",
|
|
12653
12653
|
"static": true,
|
|
@@ -12664,7 +12664,7 @@
|
|
|
12664
12664
|
"immutable": true,
|
|
12665
12665
|
"locationInModule": {
|
|
12666
12666
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12667
|
-
"line":
|
|
12667
|
+
"line": 72
|
|
12668
12668
|
},
|
|
12669
12669
|
"name": "SDK_2_21_1",
|
|
12670
12670
|
"static": true,
|
|
@@ -12681,7 +12681,7 @@
|
|
|
12681
12681
|
"immutable": true,
|
|
12682
12682
|
"locationInModule": {
|
|
12683
12683
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12684
|
-
"line":
|
|
12684
|
+
"line": 65
|
|
12685
12685
|
},
|
|
12686
12686
|
"name": "SDK_2_22_0",
|
|
12687
12687
|
"static": true,
|
|
@@ -12698,7 +12698,7 @@
|
|
|
12698
12698
|
"immutable": true,
|
|
12699
12699
|
"locationInModule": {
|
|
12700
12700
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12701
|
-
"line":
|
|
12701
|
+
"line": 58
|
|
12702
12702
|
},
|
|
12703
12703
|
"name": "SDK_2_23_0",
|
|
12704
12704
|
"static": true,
|
|
@@ -12715,7 +12715,7 @@
|
|
|
12715
12715
|
"immutable": true,
|
|
12716
12716
|
"locationInModule": {
|
|
12717
12717
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12718
|
-
"line":
|
|
12718
|
+
"line": 51
|
|
12719
12719
|
},
|
|
12720
12720
|
"name": "SDK_2_24_0",
|
|
12721
12721
|
"static": true,
|
|
@@ -12723,6 +12723,74 @@
|
|
|
12723
12723
|
"fqn": "aws-cdk-neuronx-patterns.INeuronxImage"
|
|
12724
12724
|
}
|
|
12725
12725
|
},
|
|
12726
|
+
{
|
|
12727
|
+
"const": true,
|
|
12728
|
+
"docs": {
|
|
12729
|
+
"stability": "stable",
|
|
12730
|
+
"summary": "Neuron SDK 2.25.0."
|
|
12731
|
+
},
|
|
12732
|
+
"immutable": true,
|
|
12733
|
+
"locationInModule": {
|
|
12734
|
+
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12735
|
+
"line": 44
|
|
12736
|
+
},
|
|
12737
|
+
"name": "SDK_2_25_0",
|
|
12738
|
+
"static": true,
|
|
12739
|
+
"type": {
|
|
12740
|
+
"fqn": "aws-cdk-neuronx-patterns.INeuronxImage"
|
|
12741
|
+
}
|
|
12742
|
+
},
|
|
12743
|
+
{
|
|
12744
|
+
"const": true,
|
|
12745
|
+
"docs": {
|
|
12746
|
+
"stability": "stable",
|
|
12747
|
+
"summary": "Neuron SDK 2.26.1."
|
|
12748
|
+
},
|
|
12749
|
+
"immutable": true,
|
|
12750
|
+
"locationInModule": {
|
|
12751
|
+
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12752
|
+
"line": 37
|
|
12753
|
+
},
|
|
12754
|
+
"name": "SDK_2_26_1",
|
|
12755
|
+
"static": true,
|
|
12756
|
+
"type": {
|
|
12757
|
+
"fqn": "aws-cdk-neuronx-patterns.INeuronxImage"
|
|
12758
|
+
}
|
|
12759
|
+
},
|
|
12760
|
+
{
|
|
12761
|
+
"const": true,
|
|
12762
|
+
"docs": {
|
|
12763
|
+
"stability": "stable",
|
|
12764
|
+
"summary": "Neuron SDK 2.27.1."
|
|
12765
|
+
},
|
|
12766
|
+
"immutable": true,
|
|
12767
|
+
"locationInModule": {
|
|
12768
|
+
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12769
|
+
"line": 30
|
|
12770
|
+
},
|
|
12771
|
+
"name": "SDK_2_27_1",
|
|
12772
|
+
"static": true,
|
|
12773
|
+
"type": {
|
|
12774
|
+
"fqn": "aws-cdk-neuronx-patterns.INeuronxImage"
|
|
12775
|
+
}
|
|
12776
|
+
},
|
|
12777
|
+
{
|
|
12778
|
+
"const": true,
|
|
12779
|
+
"docs": {
|
|
12780
|
+
"stability": "stable",
|
|
12781
|
+
"summary": "Neuron SDK 2.28.0."
|
|
12782
|
+
},
|
|
12783
|
+
"immutable": true,
|
|
12784
|
+
"locationInModule": {
|
|
12785
|
+
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
12786
|
+
"line": 23
|
|
12787
|
+
},
|
|
12788
|
+
"name": "SDK_2_28_0",
|
|
12789
|
+
"static": true,
|
|
12790
|
+
"type": {
|
|
12791
|
+
"fqn": "aws-cdk-neuronx-patterns.INeuronxImage"
|
|
12792
|
+
}
|
|
12793
|
+
},
|
|
12726
12794
|
{
|
|
12727
12795
|
"const": true,
|
|
12728
12796
|
"docs": {
|
|
@@ -13673,7 +13741,7 @@
|
|
|
13673
13741
|
"kind": "class",
|
|
13674
13742
|
"locationInModule": {
|
|
13675
13743
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
13676
|
-
"line":
|
|
13744
|
+
"line": 236
|
|
13677
13745
|
},
|
|
13678
13746
|
"methods": [
|
|
13679
13747
|
{
|
|
@@ -13682,7 +13750,7 @@
|
|
|
13682
13750
|
},
|
|
13683
13751
|
"locationInModule": {
|
|
13684
13752
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
13685
|
-
"line":
|
|
13753
|
+
"line": 291
|
|
13686
13754
|
},
|
|
13687
13755
|
"name": "fromNeuronSdkVersion",
|
|
13688
13756
|
"parameters": [
|
|
@@ -13730,7 +13798,7 @@
|
|
|
13730
13798
|
"immutable": true,
|
|
13731
13799
|
"locationInModule": {
|
|
13732
13800
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
13733
|
-
"line":
|
|
13801
|
+
"line": 289
|
|
13734
13802
|
},
|
|
13735
13803
|
"name": "LATEST",
|
|
13736
13804
|
"static": true,
|
|
@@ -13738,6 +13806,40 @@
|
|
|
13738
13806
|
"fqn": "aws-cdk-neuronx-patterns.IVllmInferenceNeuronxImage"
|
|
13739
13807
|
}
|
|
13740
13808
|
},
|
|
13809
|
+
{
|
|
13810
|
+
"const": true,
|
|
13811
|
+
"docs": {
|
|
13812
|
+
"stability": "stable",
|
|
13813
|
+
"summary": "Neuron SDK 2.24.1 with vLLM 0.7.2."
|
|
13814
|
+
},
|
|
13815
|
+
"immutable": true,
|
|
13816
|
+
"locationInModule": {
|
|
13817
|
+
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
13818
|
+
"line": 282
|
|
13819
|
+
},
|
|
13820
|
+
"name": "SDK_2_24_1",
|
|
13821
|
+
"static": true,
|
|
13822
|
+
"type": {
|
|
13823
|
+
"fqn": "aws-cdk-neuronx-patterns.IVllmInferenceNeuronxImage"
|
|
13824
|
+
}
|
|
13825
|
+
},
|
|
13826
|
+
{
|
|
13827
|
+
"const": true,
|
|
13828
|
+
"docs": {
|
|
13829
|
+
"stability": "stable",
|
|
13830
|
+
"summary": "Neuron SDK 2.25.0 with vLLM 0.9.1."
|
|
13831
|
+
},
|
|
13832
|
+
"immutable": true,
|
|
13833
|
+
"locationInModule": {
|
|
13834
|
+
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
13835
|
+
"line": 275
|
|
13836
|
+
},
|
|
13837
|
+
"name": "SDK_2_25_0",
|
|
13838
|
+
"static": true,
|
|
13839
|
+
"type": {
|
|
13840
|
+
"fqn": "aws-cdk-neuronx-patterns.IVllmInferenceNeuronxImage"
|
|
13841
|
+
}
|
|
13842
|
+
},
|
|
13741
13843
|
{
|
|
13742
13844
|
"const": true,
|
|
13743
13845
|
"docs": {
|
|
@@ -13747,13 +13849,81 @@
|
|
|
13747
13849
|
"immutable": true,
|
|
13748
13850
|
"locationInModule": {
|
|
13749
13851
|
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
13750
|
-
"line":
|
|
13852
|
+
"line": 268
|
|
13751
13853
|
},
|
|
13752
13854
|
"name": "SDK_2_26_0",
|
|
13753
13855
|
"static": true,
|
|
13754
13856
|
"type": {
|
|
13755
13857
|
"fqn": "aws-cdk-neuronx-patterns.IVllmInferenceNeuronxImage"
|
|
13756
13858
|
}
|
|
13859
|
+
},
|
|
13860
|
+
{
|
|
13861
|
+
"const": true,
|
|
13862
|
+
"docs": {
|
|
13863
|
+
"stability": "stable",
|
|
13864
|
+
"summary": "Neuron SDK 2.26.1 with vLLM 0.9.1."
|
|
13865
|
+
},
|
|
13866
|
+
"immutable": true,
|
|
13867
|
+
"locationInModule": {
|
|
13868
|
+
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
13869
|
+
"line": 261
|
|
13870
|
+
},
|
|
13871
|
+
"name": "SDK_2_26_1",
|
|
13872
|
+
"static": true,
|
|
13873
|
+
"type": {
|
|
13874
|
+
"fqn": "aws-cdk-neuronx-patterns.IVllmInferenceNeuronxImage"
|
|
13875
|
+
}
|
|
13876
|
+
},
|
|
13877
|
+
{
|
|
13878
|
+
"const": true,
|
|
13879
|
+
"docs": {
|
|
13880
|
+
"stability": "stable",
|
|
13881
|
+
"summary": "Neuron SDK 2.27.1 with vLLM 0.11.0."
|
|
13882
|
+
},
|
|
13883
|
+
"immutable": true,
|
|
13884
|
+
"locationInModule": {
|
|
13885
|
+
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
13886
|
+
"line": 253
|
|
13887
|
+
},
|
|
13888
|
+
"name": "SDK_2_27_1_VLLM_0_11_0",
|
|
13889
|
+
"static": true,
|
|
13890
|
+
"type": {
|
|
13891
|
+
"fqn": "aws-cdk-neuronx-patterns.IVllmInferenceNeuronxImage"
|
|
13892
|
+
}
|
|
13893
|
+
},
|
|
13894
|
+
{
|
|
13895
|
+
"const": true,
|
|
13896
|
+
"docs": {
|
|
13897
|
+
"stability": "stable",
|
|
13898
|
+
"summary": "Neuron SDK 2.27.1 with vLLM 0.13.0."
|
|
13899
|
+
},
|
|
13900
|
+
"immutable": true,
|
|
13901
|
+
"locationInModule": {
|
|
13902
|
+
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
13903
|
+
"line": 245
|
|
13904
|
+
},
|
|
13905
|
+
"name": "SDK_2_27_1_VLLM_0_13_0",
|
|
13906
|
+
"static": true,
|
|
13907
|
+
"type": {
|
|
13908
|
+
"fqn": "aws-cdk-neuronx-patterns.IVllmInferenceNeuronxImage"
|
|
13909
|
+
}
|
|
13910
|
+
},
|
|
13911
|
+
{
|
|
13912
|
+
"const": true,
|
|
13913
|
+
"docs": {
|
|
13914
|
+
"stability": "stable",
|
|
13915
|
+
"summary": "Neuron SDK 2.28.0 with vLLM 0.13.0."
|
|
13916
|
+
},
|
|
13917
|
+
"immutable": true,
|
|
13918
|
+
"locationInModule": {
|
|
13919
|
+
"filename": "src/base/neuronx/deep-learning-containers.ts",
|
|
13920
|
+
"line": 238
|
|
13921
|
+
},
|
|
13922
|
+
"name": "SDK_2_28_0",
|
|
13923
|
+
"static": true,
|
|
13924
|
+
"type": {
|
|
13925
|
+
"fqn": "aws-cdk-neuronx-patterns.IVllmInferenceNeuronxImage"
|
|
13926
|
+
}
|
|
13757
13927
|
}
|
|
13758
13928
|
],
|
|
13759
13929
|
"symbolId": "src/base/neuronx/deep-learning-containers:VllmInferenceNeuronxImage"
|
|
@@ -15672,7 +15842,7 @@
|
|
|
15672
15842
|
},
|
|
15673
15843
|
"locationInModule": {
|
|
15674
15844
|
"filename": "src/vllm-nxd-inference/vllm-nxd-inference-compiler.ts",
|
|
15675
|
-
"line":
|
|
15845
|
+
"line": 43
|
|
15676
15846
|
},
|
|
15677
15847
|
"parameters": [
|
|
15678
15848
|
{
|
|
@@ -15699,7 +15869,7 @@
|
|
|
15699
15869
|
"kind": "class",
|
|
15700
15870
|
"locationInModule": {
|
|
15701
15871
|
"filename": "src/vllm-nxd-inference/vllm-nxd-inference-compiler.ts",
|
|
15702
|
-
"line":
|
|
15872
|
+
"line": 38
|
|
15703
15873
|
},
|
|
15704
15874
|
"name": "VllmNxdInferenceCompileImage",
|
|
15705
15875
|
"properties": [
|
|
@@ -15711,7 +15881,7 @@
|
|
|
15711
15881
|
"immutable": true,
|
|
15712
15882
|
"locationInModule": {
|
|
15713
15883
|
"filename": "src/vllm-nxd-inference/vllm-nxd-inference-compiler.ts",
|
|
15714
|
-
"line":
|
|
15884
|
+
"line": 42
|
|
15715
15885
|
},
|
|
15716
15886
|
"name": "image",
|
|
15717
15887
|
"overrides": "aws-cdk-neuronx-patterns.VllmNxdInferenceEcsImageBase",
|
|
@@ -15733,7 +15903,7 @@
|
|
|
15733
15903
|
"kind": "interface",
|
|
15734
15904
|
"locationInModule": {
|
|
15735
15905
|
"filename": "src/vllm-nxd-inference/vllm-nxd-inference-compiler.ts",
|
|
15736
|
-
"line":
|
|
15906
|
+
"line": 64
|
|
15737
15907
|
},
|
|
15738
15908
|
"name": "VllmNxdInferenceCompileProps",
|
|
15739
15909
|
"properties": [
|
|
@@ -15746,7 +15916,7 @@
|
|
|
15746
15916
|
"immutable": true,
|
|
15747
15917
|
"locationInModule": {
|
|
15748
15918
|
"filename": "src/vllm-nxd-inference/vllm-nxd-inference-compiler.ts",
|
|
15749
|
-
"line":
|
|
15919
|
+
"line": 72
|
|
15750
15920
|
},
|
|
15751
15921
|
"name": "bucket",
|
|
15752
15922
|
"type": {
|
|
@@ -15762,7 +15932,7 @@
|
|
|
15762
15932
|
"immutable": true,
|
|
15763
15933
|
"locationInModule": {
|
|
15764
15934
|
"filename": "src/vllm-nxd-inference/vllm-nxd-inference-compiler.ts",
|
|
15765
|
-
"line":
|
|
15935
|
+
"line": 80
|
|
15766
15936
|
},
|
|
15767
15937
|
"name": "model",
|
|
15768
15938
|
"type": {
|
|
@@ -15778,7 +15948,7 @@
|
|
|
15778
15948
|
"immutable": true,
|
|
15779
15949
|
"locationInModule": {
|
|
15780
15950
|
"filename": "src/vllm-nxd-inference/vllm-nxd-inference-compiler.ts",
|
|
15781
|
-
"line":
|
|
15951
|
+
"line": 68
|
|
15782
15952
|
},
|
|
15783
15953
|
"name": "vpc",
|
|
15784
15954
|
"type": {
|
|
@@ -15796,7 +15966,7 @@
|
|
|
15796
15966
|
"immutable": true,
|
|
15797
15967
|
"locationInModule": {
|
|
15798
15968
|
"filename": "src/vllm-nxd-inference/vllm-nxd-inference-compiler.ts",
|
|
15799
|
-
"line":
|
|
15969
|
+
"line": 104
|
|
15800
15970
|
},
|
|
15801
15971
|
"name": "environment",
|
|
15802
15972
|
"optional": true,
|
|
@@ -15819,7 +15989,7 @@
|
|
|
15819
15989
|
"immutable": true,
|
|
15820
15990
|
"locationInModule": {
|
|
15821
15991
|
"filename": "src/vllm-nxd-inference/vllm-nxd-inference-compiler.ts",
|
|
15822
|
-
"line":
|
|
15992
|
+
"line": 116
|
|
15823
15993
|
},
|
|
15824
15994
|
"name": "image",
|
|
15825
15995
|
"optional": true,
|
|
@@ -15836,7 +16006,7 @@
|
|
|
15836
16006
|
"immutable": true,
|
|
15837
16007
|
"locationInModule": {
|
|
15838
16008
|
"filename": "src/vllm-nxd-inference/vllm-nxd-inference-compiler.ts",
|
|
15839
|
-
"line":
|
|
16009
|
+
"line": 76
|
|
15840
16010
|
},
|
|
15841
16011
|
"name": "neuronxInstanceType",
|
|
15842
16012
|
"optional": true,
|
|
@@ -15855,7 +16025,7 @@
|
|
|
15855
16025
|
"immutable": true,
|
|
15856
16026
|
"locationInModule": {
|
|
15857
16027
|
"filename": "src/vllm-nxd-inference/vllm-nxd-inference-compiler.ts",
|
|
15858
|
-
"line":
|
|
16028
|
+
"line": 91
|
|
15859
16029
|
},
|
|
15860
16030
|
"name": "spot",
|
|
15861
16031
|
"optional": true,
|
|
@@ -15873,7 +16043,7 @@
|
|
|
15873
16043
|
"immutable": true,
|
|
15874
16044
|
"locationInModule": {
|
|
15875
16045
|
"filename": "src/vllm-nxd-inference/vllm-nxd-inference-compiler.ts",
|
|
15876
|
-
"line":
|
|
16046
|
+
"line": 111
|
|
15877
16047
|
},
|
|
15878
16048
|
"name": "vllmArgs",
|
|
15879
16049
|
"optional": true,
|
|
@@ -15891,7 +16061,7 @@
|
|
|
15891
16061
|
"immutable": true,
|
|
15892
16062
|
"locationInModule": {
|
|
15893
16063
|
"filename": "src/vllm-nxd-inference/vllm-nxd-inference-compiler.ts",
|
|
15894
|
-
"line":
|
|
16064
|
+
"line": 85
|
|
15895
16065
|
},
|
|
15896
16066
|
"name": "volumeSize",
|
|
15897
16067
|
"optional": true,
|
|
@@ -15909,7 +16079,7 @@
|
|
|
15909
16079
|
"immutable": true,
|
|
15910
16080
|
"locationInModule": {
|
|
15911
16081
|
"filename": "src/vllm-nxd-inference/vllm-nxd-inference-compiler.ts",
|
|
15912
|
-
"line":
|
|
16082
|
+
"line": 97
|
|
15913
16083
|
},
|
|
15914
16084
|
"name": "vpcSubnets",
|
|
15915
16085
|
"optional": true,
|
|
@@ -15934,7 +16104,7 @@
|
|
|
15934
16104
|
"kind": "interface",
|
|
15935
16105
|
"locationInModule": {
|
|
15936
16106
|
"filename": "src/vllm-nxd-inference/vllm-nxd-inference-compiler.ts",
|
|
15937
|
-
"line":
|
|
16107
|
+
"line": 122
|
|
15938
16108
|
},
|
|
15939
16109
|
"name": "VllmNxdInferenceCompiledModel",
|
|
15940
16110
|
"properties": [
|
|
@@ -15947,7 +16117,7 @@
|
|
|
15947
16117
|
"immutable": true,
|
|
15948
16118
|
"locationInModule": {
|
|
15949
16119
|
"filename": "src/vllm-nxd-inference/vllm-nxd-inference-compiler.ts",
|
|
15950
|
-
"line":
|
|
16120
|
+
"line": 126
|
|
15951
16121
|
},
|
|
15952
16122
|
"name": "vllmArgs",
|
|
15953
16123
|
"type": {
|
|
@@ -15972,7 +16142,7 @@
|
|
|
15972
16142
|
},
|
|
15973
16143
|
"locationInModule": {
|
|
15974
16144
|
"filename": "src/vllm-nxd-inference/vllm-nxd-inference-compiler.ts",
|
|
15975
|
-
"line":
|
|
16145
|
+
"line": 136
|
|
15976
16146
|
},
|
|
15977
16147
|
"parameters": [
|
|
15978
16148
|
{
|
|
@@ -15998,7 +16168,7 @@
|
|
|
15998
16168
|
"kind": "class",
|
|
15999
16169
|
"locationInModule": {
|
|
16000
16170
|
"filename": "src/vllm-nxd-inference/vllm-nxd-inference-compiler.ts",
|
|
16001
|
-
"line":
|
|
16171
|
+
"line": 133
|
|
16002
16172
|
},
|
|
16003
16173
|
"methods": [
|
|
16004
16174
|
{
|
|
@@ -16009,7 +16179,7 @@
|
|
|
16009
16179
|
},
|
|
16010
16180
|
"locationInModule": {
|
|
16011
16181
|
"filename": "src/vllm-nxd-inference/vllm-nxd-inference-compiler.ts",
|
|
16012
|
-
"line":
|
|
16182
|
+
"line": 251
|
|
16013
16183
|
},
|
|
16014
16184
|
"name": "compile",
|
|
16015
16185
|
"returns": {
|
|
@@ -16919,6 +17089,6 @@
|
|
|
16919
17089
|
"symbolId": "src/base/server-engine/vllm-engine/vllm-engine-argments:VllmTask"
|
|
16920
17090
|
}
|
|
16921
17091
|
},
|
|
16922
|
-
"version": "0.
|
|
16923
|
-
"fingerprint": "
|
|
17092
|
+
"version": "0.2.0",
|
|
17093
|
+
"fingerprint": "y3L7pFC63WvgygoXeexNy9PC78TL0KjiDkaYJ1KI0l8="
|
|
16924
17094
|
}
|