Skip to content

Commit 42c344b

Browse files
pintaoz-awspintaoz-aws
authored
Merge branch 'master' into inference-spec
2 parents c966560 + 143c128 commit 42c344b

File tree

16 files changed

+693
-122
lines changed

16 files changed

+693
-122
lines changed

‎CHANGELOG.md‎

Lines changed: 16 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,21 @@
11
# Changelog
22

3+
## v2.252.0 (2025-09-29)
4+
5+
### Features
6+
7+
* change S3 endpoint env name
8+
* add eval custom lambda arn to hyperparameters
9+
10+
### Bug Fixes and Other Changes
11+
12+
* merge rba without the iso region changes
13+
* handle trial component status message longer than API supports
14+
* Add nova custom lambda in hyperparameter from estimator
15+
* add retryable option to emr step in SageMaker Pipelines
16+
* Feature/js mlops telemetry
17+
* latest tgi
18+
319
## v2.251.1 (2025-08-29)
420

521
### Bug Fixes and Other Changes

‎VERSION‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1 +1 @@
1-
2.251.2.dev0
1+
2.252.1.dev0

‎src/sagemaker/experiments/run.py‎

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -68,6 +68,7 @@
6868
TRIAL_NAME_TEMPLATE = "Default-Run-Group-{}"
6969
MAX_RUN_TC_ARTIFACTS_LEN = 30
7070
MAX_NAME_LEN_IN_BACKEND = 120
71+
MAX_STATUS_MESSAGE_LEN = 1024
7172
EXPERIMENT_NAME = "ExperimentName"
7273
TRIAL_NAME = "TrialName"
7374
RUN_NAME = "RunName"
@@ -759,7 +760,7 @@ def __exit__(self, exc_type, exc_value, exc_traceback):
759760
if exc_value:
760761
self._trial_component.status = _api_types.TrialComponentStatus(
761762
primary_status=_TrialComponentStatusType.Failed.value,
762-
message=str(exc_value),
763+
message=(str(exc_value) or "")[:MAX_STATUS_MESSAGE_LEN],
763764
)
764765
else:
765766
self._trial_component.status = _api_types.TrialComponentStatus(

‎src/sagemaker/image_uri_config/huggingface-llm-neuronx.json‎

Lines changed: 1 addition & 56 deletions
Original file line numberDiff line numberDiff line change
@@ -4,8 +4,7 @@
44
"inf2"
55
],
66
"version_aliases": {
7-
"0.0": "0.0.28",
8-
"0.2": "0.2.0"
7+
"0.0": "0.0.28"
98
},
109
"versions": {
1110
"0.0.16": {
@@ -655,60 +654,6 @@
655654
"container_version": {
656655
"inf2": "ubuntu22.04"
657656
}
658-
},
659-
"0.2.0": {
660-
"py_versions": [
661-
"py310"
662-
],
663-
"registries": {
664-
"af-south-1": "626614931356",
665-
"ap-east-1": "871362719292",
666-
"ap-east-2": "975050140332",
667-
"ap-northeast-1": "763104351884",
668-
"ap-northeast-2": "763104351884",
669-
"ap-northeast-3": "364406365360",
670-
"ap-south-1": "763104351884",
671-
"ap-south-2": "772153158452",
672-
"ap-southeast-1": "763104351884",
673-
"ap-southeast-2": "763104351884",
674-
"ap-southeast-3": "907027046896",
675-
"ap-southeast-4": "457447274322",
676-
"ap-southeast-5": "550225433462",
677-
"ap-southeast-6": "633930458069",
678-
"ap-southeast-7": "590183813437",
679-
"ca-central-1": "763104351884",
680-
"ca-west-1": "204538143572",
681-
"cn-north-1": "727897471807",
682-
"cn-northwest-1": "727897471807",
683-
"eu-central-1": "763104351884",
684-
"eu-central-2": "380420809688",
685-
"eu-north-1": "763104351884",
686-
"eu-south-1": "692866216735",
687-
"eu-south-2": "503227376785",
688-
"eu-west-1": "763104351884",
689-
"eu-west-2": "763104351884",
690-
"eu-west-3": "763104351884",
691-
"il-central-1": "780543022126",
692-
"me-central-1": "914824155844",
693-
"me-south-1": "217643126080",
694-
"mx-central-1": "637423239942",
695-
"sa-east-1": "763104351884",
696-
"us-east-1": "763104351884",
697-
"us-east-2": "763104351884",
698-
"us-gov-east-1": "446045086412",
699-
"us-gov-west-1": "442386744353",
700-
"us-iso-east-1": "886529160074",
701-
"us-isob-east-1": "094389454867",
702-
"us-isof-east-1": "303241398832",
703-
"us-isof-south-1": "454834333376",
704-
"us-west-1": "763104351884",
705-
"us-west-2": "763104351884"
706-
},
707-
"tag_prefix": "2.5.1-optimum3.3.4",
708-
"repository": "huggingface-pytorch-tgi-inference",
709-
"container_version": {
710-
"inf2": "ubuntu22.04"
711-
}
712657
}
713658
}
714659
}

‎src/sagemaker/image_uri_config/huggingface-llm.json‎

Lines changed: 1 addition & 55 deletions
Original file line numberDiff line numberDiff line change
@@ -16,8 +16,7 @@
1616
"2.3": "2.3.1",
1717
"3.0": "3.0.1",
1818
"3.2": "3.2.3",
19-
"3.1": "3.1.1",
20-
"3.3": "3.3.4"
19+
"3.1": "3.1.1"
2120
},
2221
"versions": {
2322
"0.6.0": {
@@ -1153,59 +1152,6 @@
11531152
"container_version": {
11541153
"gpu": "cu124-ubuntu22.04"
11551154
}
1156-
},
1157-
"3.3.4": {
1158-
"py_versions": [
1159-
"py311"
1160-
],
1161-
"registries": {
1162-
"af-south-1": "626614931356",
1163-
"ap-east-1": "871362719292",
1164-
"ap-east-2": "975050140332",
1165-
"ap-northeast-1": "763104351884",
1166-
"ap-northeast-2": "763104351884",
1167-
"ap-northeast-3": "364406365360",
1168-
"ap-south-1": "763104351884",
1169-
"ap-south-2": "772153158452",
1170-
"ap-southeast-1": "763104351884",
1171-
"ap-southeast-2": "763104351884",
1172-
"ap-southeast-3": "907027046896",
1173-
"ap-southeast-4": "457447274322",
1174-
"ap-southeast-5": "550225433462",
1175-
"ap-southeast-7": "590183813437",
1176-
"ca-central-1": "763104351884",
1177-
"ca-west-1": "204538143572",
1178-
"cn-north-1": "727897471807",
1179-
"cn-northwest-1": "727897471807",
1180-
"eu-central-1": "763104351884",
1181-
"eu-central-2": "380420809688",
1182-
"eu-north-1": "763104351884",
1183-
"eu-south-1": "692866216735",
1184-
"eu-south-2": "503227376785",
1185-
"eu-west-1": "763104351884",
1186-
"eu-west-2": "763104351884",
1187-
"eu-west-3": "763104351884",
1188-
"il-central-1": "780543022126",
1189-
"me-central-1": "914824155844",
1190-
"me-south-1": "217643126080",
1191-
"mx-central-1": "637423239942",
1192-
"sa-east-1": "763104351884",
1193-
"us-east-1": "763104351884",
1194-
"us-east-2": "763104351884",
1195-
"us-gov-east-1": "446045086412",
1196-
"us-gov-west-1": "442386744353",
1197-
"us-iso-east-1": "886529160074",
1198-
"us-isob-east-1": "094389454867",
1199-
"us-isof-east-1": "303241398832",
1200-
"us-isof-south-1": "454834333376",
1201-
"us-west-1": "763104351884",
1202-
"us-west-2": "763104351884"
1203-
},
1204-
"tag_prefix": "2.7.0-tgi3.3.4",
1205-
"repository": "huggingface-pytorch-tgi-inference",
1206-
"container_version": {
1207-
"gpu": "cu124-ubuntu22.04"
1208-
}
12091155
}
12101156
}
12111157
}

‎src/sagemaker/image_uri_config/pytorch.json‎

Lines changed: 48 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1737,7 +1737,8 @@
17371737
"2.4": "2.4.0",
17381738
"2.5": "2.5.1",
17391739
"2.6": "2.6.0",
1740-
"2.7": "2.7.1"
1740+
"2.7": "2.7.1",
1741+
"2.8": "2.8.0"
17411742
},
17421743
"versions": {
17431744
"0.4.0": {
@@ -3048,6 +3049,52 @@
30483049
"us-west-2": "763104351884"
30493050
},
30503051
"repository": "pytorch-training"
3052+
},
3053+
"2.8.0": {
3054+
"py_versions": [
3055+
"py312"
3056+
],
3057+
"registries": {
3058+
"af-south-1": "626614931356",
3059+
"ap-east-1": "871362719292",
3060+
"ap-east-2": "975050140332",
3061+
"ap-northeast-1": "763104351884",
3062+
"ap-northeast-2": "763104351884",
3063+
"ap-northeast-3": "364406365360",
3064+
"ap-south-1": "763104351884",
3065+
"ap-south-2": "772153158452",
3066+
"ap-southeast-1": "763104351884",
3067+
"ap-southeast-2": "763104351884",
3068+
"ap-southeast-3": "907027046896",
3069+
"ap-southeast-4": "457447274322",
3070+
"ap-southeast-5": "550225433462",
3071+
"ap-southeast-6": "633930458069",
3072+
"ap-southeast-7": "590183813437",
3073+
"ca-central-1": "763104351884",
3074+
"ca-west-1": "204538143572",
3075+
"cn-north-1": "727897471807",
3076+
"cn-northwest-1": "727897471807",
3077+
"eu-central-1": "763104351884",
3078+
"eu-central-2": "380420809688",
3079+
"eu-north-1": "763104351884",
3080+
"eu-south-1": "692866216735",
3081+
"eu-south-2": "503227376785",
3082+
"eu-west-1": "763104351884",
3083+
"eu-west-2": "763104351884",
3084+
"eu-west-3": "763104351884",
3085+
"il-central-1": "780543022126",
3086+
"me-central-1": "914824155844",
3087+
"me-south-1": "217643126080",
3088+
"mx-central-1": "637423239942",
3089+
"sa-east-1": "763104351884",
3090+
"us-east-1": "763104351884",
3091+
"us-east-2": "763104351884",
3092+
"us-gov-east-1": "446045086412",
3093+
"us-gov-west-1": "442386744353",
3094+
"us-west-1": "763104351884",
3095+
"us-west-2": "763104351884"
3096+
},
3097+
"repository": "pytorch-training"
30513098
}
30523099
}
30533100
}

‎src/sagemaker/jumpstart/region_config.json‎

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -57,6 +57,10 @@
5757
"content_bucket": "jumpstart-cache-prod-ap-southeast-5",
5858
"gated_content_bucket": "jumpstart-private-cache-prod-ap-southeast-5"
5959
},
60+
"ap-southeast-6": {
61+
"content_bucket": "jumpstart-cache-prod-ap-southeast-6",
62+
"gated_content_bucket": "jumpstart-private-cache-prod-ap-southeast-6"
63+
},
6064
"ap-southeast-7": {
6165
"content_bucket": "jumpstart-cache-prod-ap-southeast-7",
6266
"gated_content_bucket": "jumpstart-private-cache-prod-ap-southeast-7"

‎src/sagemaker/local/image.py‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -50,7 +50,7 @@
5050
# Environment variables to be set during training
5151
REGION_ENV_NAME = "AWS_REGION"
5252
TRAINING_JOB_NAME_ENV_NAME = "TRAINING_JOB_NAME"
53-
S3_ENDPOINT_URL_ENV_NAME = "S3_ENDPOINT_URL"
53+
S3_ENDPOINT_URL_ENV_NAME = "AWS_ENDPOINT_URL_S3"
5454
SM_STUDIO_LOCAL_MODE = "SM_STUDIO_LOCAL_MODE"
5555

5656
# SELinux Enabled

‎src/sagemaker/modules/train/sm_recipes/utils.py‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -310,7 +310,7 @@ def _get_args_from_nova_recipe(
310310
processor = recipe.get("processor", {})
311311
lambda_arn = processor.get("lambda_arn", "")
312312
if lambda_arn:
313-
args["hyperparameters"]["lambda_arn"] = lambda_arn
313+
args["hyperparameters"]["eval_lambda_arn"] = lambda_arn
314314

315315
_register_custom_resolvers()
316316

‎src/sagemaker/pytorch/estimator.py‎

Lines changed: 7 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1224,6 +1224,13 @@ def _setup_for_nova_recipe(
12241224
)
12251225
args["hyperparameters"]["kms_key"] = kms_key
12261226

1227+
# Handle eval custom lambda configuration
1228+
if recipe.get("evaluation", {}):
1229+
processor = recipe.get("processor", {})
1230+
lambda_arn = processor.get("lambda_arn", "")
1231+
if lambda_arn:
1232+
args["hyperparameters"]["eval_lambda_arn"] = lambda_arn
1233+
12271234
# Resolve and save the final recipe
12281235
self._recipe_resolve_and_save(recipe, recipe_name, args["source_dir"])
12291236

0 commit comments

Comments
 (0)