Skip to content

Commit e070c15

Browse files
beniericbenieric
authored
Merge branch 'master' into master
2 parents 0cb1caa + 6d2dfa0 commit e070c15

35 files changed

+2018
-87
lines changed

‎CHANGELOG.md

+23
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,28 @@
11
# Changelog
22

3+
## v2.238.0 (2025-01-29)
4+
5+
### Features
6+
7+
* use jumpstart deployment config image as default optimization image
8+
9+
### Bug Fixes and Other Changes
10+
11+
* chore: add new images for HF TGI
12+
* update image_uri_configs 01-29-2025 06:18:08 PST
13+
* skip TF tests for unsupported versions
14+
* Merge branch 'master-rba' into local_merge
15+
* Add missing attributes to local resourceconfig
16+
* update image_uri_configs 01-27-2025 06:18:13 PST
17+
* update image_uri_configs 01-24-2025 06:18:11 PST
18+
* add missing schema definition in docs
19+
* Omegaconf upgrade
20+
* SageMaker @remote function: Added multi-node functionality
21+
* remove option
22+
* fix typo
23+
* fix tests
24+
* Add an option for user to remove inputs and container artifacts when using local model trainer
25+
326
## v2.237.3 (2025-01-09)
427

528
### Bug Fixes and Other Changes

‎VERSION

+1-1
Original file line numberDiff line numberDiff line change
@@ -1 +1 @@
1-
2.237.4.dev0
1+
2.238.1.dev0

‎src/sagemaker/fw_utils.py

+1
Original file line numberDiff line numberDiff line change
@@ -152,6 +152,7 @@
152152
"2.1.0",
153153
"2.1.2",
154154
"2.2.0",
155+
"2.3.0",
155156
"2.3.1",
156157
"2.4.1",
157158
]

‎src/sagemaker/image_uri_config/huggingface-llm.json

+96-1
Original file line numberDiff line numberDiff line change
@@ -12,7 +12,8 @@
1212
"1.2": "1.2.0",
1313
"1.3": "1.3.3",
1414
"1.4": "1.4.5",
15-
"2.0": "2.3.1"
15+
"2.0": "2.4.0",
16+
"3.0": "3.0.1"
1617
},
1718
"versions": {
1819
"0.6.0": {
@@ -766,6 +767,100 @@
766767
"container_version": {
767768
"gpu": "cu124-ubuntu22.04"
768769
}
770+
},
771+
"2.4.0": {
772+
"py_versions": [
773+
"py311"
774+
],
775+
"registries": {
776+
"af-south-1": "626614931356",
777+
"il-central-1": "780543022126",
778+
"ap-east-1": "871362719292",
779+
"ap-northeast-1": "763104351884",
780+
"ap-northeast-2": "763104351884",
781+
"ap-northeast-3": "364406365360",
782+
"ap-south-1": "763104351884",
783+
"ap-south-2": "772153158452",
784+
"ap-southeast-1": "763104351884",
785+
"ap-southeast-2": "763104351884",
786+
"ap-southeast-3": "907027046896",
787+
"ap-southeast-4": "457447274322",
788+
"ca-central-1": "763104351884",
789+
"cn-north-1": "727897471807",
790+
"cn-northwest-1": "727897471807",
791+
"eu-central-1": "763104351884",
792+
"eu-central-2": "380420809688",
793+
"eu-north-1": "763104351884",
794+
"eu-west-1": "763104351884",
795+
"eu-west-2": "763104351884",
796+
"eu-west-3": "763104351884",
797+
"eu-south-1": "692866216735",
798+
"eu-south-2": "503227376785",
799+
"me-south-1": "217643126080",
800+
"me-central-1": "914824155844",
801+
"sa-east-1": "763104351884",
802+
"us-east-1": "763104351884",
803+
"us-east-2": "763104351884",
804+
"us-gov-east-1": "446045086412",
805+
"us-gov-west-1": "442386744353",
806+
"us-iso-east-1": "886529160074",
807+
"us-isob-east-1": "094389454867",
808+
"us-west-1": "763104351884",
809+
"us-west-2": "763104351884",
810+
"ca-west-1": "204538143572"
811+
},
812+
"tag_prefix": "2.4.0-tgi2.4.0",
813+
"repository": "huggingface-pytorch-tgi-inference",
814+
"container_version": {
815+
"gpu": "cu124-ubuntu22.04-v2.2"
816+
}
817+
},
818+
"3.0.1": {
819+
"py_versions": [
820+
"py311"
821+
],
822+
"registries": {
823+
"af-south-1": "626614931356",
824+
"il-central-1": "780543022126",
825+
"ap-east-1": "871362719292",
826+
"ap-northeast-1": "763104351884",
827+
"ap-northeast-2": "763104351884",
828+
"ap-northeast-3": "364406365360",
829+
"ap-south-1": "763104351884",
830+
"ap-south-2": "772153158452",
831+
"ap-southeast-1": "763104351884",
832+
"ap-southeast-2": "763104351884",
833+
"ap-southeast-3": "907027046896",
834+
"ap-southeast-4": "457447274322",
835+
"ca-central-1": "763104351884",
836+
"cn-north-1": "727897471807",
837+
"cn-northwest-1": "727897471807",
838+
"eu-central-1": "763104351884",
839+
"eu-central-2": "380420809688",
840+
"eu-north-1": "763104351884",
841+
"eu-west-1": "763104351884",
842+
"eu-west-2": "763104351884",
843+
"eu-west-3": "763104351884",
844+
"eu-south-1": "692866216735",
845+
"eu-south-2": "503227376785",
846+
"me-south-1": "217643126080",
847+
"me-central-1": "914824155844",
848+
"sa-east-1": "763104351884",
849+
"us-east-1": "763104351884",
850+
"us-east-2": "763104351884",
851+
"us-gov-east-1": "446045086412",
852+
"us-gov-west-1": "442386744353",
853+
"us-iso-east-1": "886529160074",
854+
"us-isob-east-1": "094389454867",
855+
"us-west-1": "763104351884",
856+
"us-west-2": "763104351884",
857+
"ca-west-1": "204538143572"
858+
},
859+
"tag_prefix": "2.4.0-tgi3.0.1",
860+
"repository": "huggingface-pytorch-tgi-inference",
861+
"container_version": {
862+
"gpu": "cu124-ubuntu22.04-v2.1"
863+
}
769864
}
770865
}
771866
}

‎src/sagemaker/image_uri_config/huggingface.json

+49-1
Original file line numberDiff line numberDiff line change
@@ -13,7 +13,8 @@
1313
"4.17": "4.17.0",
1414
"4.26": "4.26.0",
1515
"4.28": "4.28.1",
16-
"4.36": "4.36.0"
16+
"4.36": "4.36.0",
17+
"4.46": "4.46.1"
1718
},
1819
"versions": {
1920
"4.4.2": {
@@ -1018,6 +1019,53 @@
10181019
"gpu": "cu121-ubuntu20.04"
10191020
}
10201021
}
1022+
},
1023+
"4.46.1": {
1024+
"version_aliases": {
1025+
"pytorch2.3": "pytorch2.3.0"
1026+
},
1027+
"pytorch2.3.0": {
1028+
"py_versions": [
1029+
"py311"
1030+
],
1031+
"registries": {
1032+
"af-south-1": "626614931356",
1033+
"il-central-1": "780543022126",
1034+
"ap-east-1": "871362719292",
1035+
"ap-northeast-1": "763104351884",
1036+
"ap-northeast-2": "763104351884",
1037+
"ap-northeast-3": "364406365360",
1038+
"ap-south-1": "763104351884",
1039+
"ap-southeast-1": "763104351884",
1040+
"ap-southeast-2": "763104351884",
1041+
"ap-southeast-3": "907027046896",
1042+
"ca-central-1": "763104351884",
1043+
"cn-north-1": "727897471807",
1044+
"cn-northwest-1": "727897471807",
1045+
"eu-central-1": "763104351884",
1046+
"eu-north-1": "763104351884",
1047+
"eu-west-1": "763104351884",
1048+
"eu-west-2": "763104351884",
1049+
"eu-west-3": "763104351884",
1050+
"eu-south-1": "692866216735",
1051+
"me-south-1": "217643126080",
1052+
"me-central-1": "914824155844",
1053+
"sa-east-1": "763104351884",
1054+
"us-east-1": "763104351884",
1055+
"us-east-2": "763104351884",
1056+
"us-gov-east-1": "446045086412",
1057+
"us-gov-west-1": "442386744353",
1058+
"us-iso-east-1": "886529160074",
1059+
"us-isob-east-1": "094389454867",
1060+
"us-west-1": "763104351884",
1061+
"us-west-2": "763104351884",
1062+
"ca-west-1": "204538143572"
1063+
},
1064+
"repository": "huggingface-pytorch-training",
1065+
"container_version": {
1066+
"gpu": "cu121-ubuntu20.04"
1067+
}
1068+
}
10211069
}
10221070
}
10231071
},

‎src/sagemaker/image_uri_config/sagemaker-base-python.json

+1
Original file line numberDiff line numberDiff line change
@@ -11,6 +11,7 @@
1111
"ap-southeast-1": "492261229750",
1212
"ap-southeast-2": "452832661640",
1313
"ap-southeast-3": "276181064229",
14+
"ap-southeast-5": "148761635175",
1415
"ca-central-1": "310906938811",
1516
"cn-north-1": "390048526115",
1617
"cn-northwest-1": "390780980154",

‎src/sagemaker/image_uri_config/tensorflow.json

+85-2
Original file line numberDiff line numberDiff line change
@@ -332,7 +332,8 @@
332332
"2.12": "2.12.1",
333333
"2.13": "2.13.0",
334334
"2.14": "2.14.1",
335-
"2.16": "2.16.1"
335+
"2.16": "2.16.1",
336+
"2.18": "2.18.0"
336337
},
337338
"versions": {
338339
"1.4.1": {
@@ -2267,6 +2268,45 @@
22672268
"us-west-2": "763104351884"
22682269
},
22692270
"repository": "tensorflow-inference"
2271+
},
2272+
"2.18.0": {
2273+
"registries": {
2274+
"af-south-1": "626614931356",
2275+
"ap-east-1": "871362719292",
2276+
"ap-northeast-1": "763104351884",
2277+
"ap-northeast-2": "763104351884",
2278+
"ap-northeast-3": "364406365360",
2279+
"ap-south-1": "763104351884",
2280+
"ap-south-2": "772153158452",
2281+
"ap-southeast-1": "763104351884",
2282+
"ap-southeast-2": "763104351884",
2283+
"ap-southeast-3": "907027046896",
2284+
"ap-southeast-4": "457447274322",
2285+
"ap-southeast-5": "550225433462",
2286+
"ca-central-1": "763104351884",
2287+
"ca-west-1": "204538143572",
2288+
"cn-north-1": "727897471807",
2289+
"cn-northwest-1": "727897471807",
2290+
"eu-central-1": "763104351884",
2291+
"eu-central-2": "380420809688",
2292+
"eu-north-1": "763104351884",
2293+
"eu-south-1": "692866216735",
2294+
"eu-south-2": "503227376785",
2295+
"eu-west-1": "763104351884",
2296+
"eu-west-2": "763104351884",
2297+
"eu-west-3": "763104351884",
2298+
"il-central-1": "780543022126",
2299+
"me-central-1": "914824155844",
2300+
"me-south-1": "217643126080",
2301+
"sa-east-1": "763104351884",
2302+
"us-east-1": "763104351884",
2303+
"us-east-2": "763104351884",
2304+
"us-gov-east-1": "446045086412",
2305+
"us-gov-west-1": "442386744353",
2306+
"us-west-1": "763104351884",
2307+
"us-west-2": "763104351884"
2308+
},
2309+
"repository": "tensorflow-inference"
22702310
}
22712311
}
22722312
},
@@ -2550,7 +2590,8 @@
25502590
"2.12": "2.12.0",
25512591
"2.13": "2.13.0",
25522592
"2.14": "2.14.1",
2553-
"2.16": "2.16.2"
2593+
"2.16": "2.16.2",
2594+
"2.18": "2.18.0"
25542595
},
25552596
"versions": {
25562597
"1.4.1": {
@@ -4570,6 +4611,48 @@
45704611
"us-west-2": "763104351884"
45714612
},
45724613
"repository": "tensorflow-training"
4614+
},
4615+
"2.18.0": {
4616+
"py_versions": [
4617+
"py310"
4618+
],
4619+
"registries": {
4620+
"af-south-1": "626614931356",
4621+
"ap-east-1": "871362719292",
4622+
"ap-northeast-1": "763104351884",
4623+
"ap-northeast-2": "763104351884",
4624+
"ap-northeast-3": "364406365360",
4625+
"ap-south-1": "763104351884",
4626+
"ap-south-2": "772153158452",
4627+
"ap-southeast-1": "763104351884",
4628+
"ap-southeast-2": "763104351884",
4629+
"ap-southeast-3": "907027046896",
4630+
"ap-southeast-4": "457447274322",
4631+
"ap-southeast-5": "550225433462",
4632+
"ca-central-1": "763104351884",
4633+
"ca-west-1": "204538143572",
4634+
"cn-north-1": "727897471807",
4635+
"cn-northwest-1": "727897471807",
4636+
"eu-central-1": "763104351884",
4637+
"eu-central-2": "380420809688",
4638+
"eu-north-1": "763104351884",
4639+
"eu-south-1": "692866216735",
4640+
"eu-south-2": "503227376785",
4641+
"eu-west-1": "763104351884",
4642+
"eu-west-2": "763104351884",
4643+
"eu-west-3": "763104351884",
4644+
"il-central-1": "780543022126",
4645+
"me-central-1": "914824155844",
4646+
"me-south-1": "217643126080",
4647+
"sa-east-1": "763104351884",
4648+
"us-east-1": "763104351884",
4649+
"us-east-2": "763104351884",
4650+
"us-gov-east-1": "446045086412",
4651+
"us-gov-west-1": "442386744353",
4652+
"us-west-1": "763104351884",
4653+
"us-west-2": "763104351884"
4654+
},
4655+
"repository": "tensorflow-training"
45734656
}
45744657
}
45754658
}

‎src/sagemaker/local/image.py

+12-2
Original file line numberDiff line numberDiff line change
@@ -473,7 +473,12 @@ def write_processing_config_files(
473473
"""
474474
config_path = os.path.join(self.container_root, host, "config")
475475

476-
resource_config = {"current_host": host, "hosts": self.hosts}
476+
resource_config = {
477+
"current_host": host,
478+
"hosts": self.hosts,
479+
"network_interface_name": "eth0",
480+
"current_instance_type": self.instance_type,
481+
}
477482
_write_json_file(os.path.join(config_path, "resourceconfig.json"), resource_config)
478483

479484
processing_job_config = {
@@ -519,7 +524,12 @@ def write_config_files(self, host, hyperparameters, input_data_config):
519524
"""
520525
config_path = os.path.join(self.container_root, host, "input", "config")
521526

522-
resource_config = {"current_host": host, "hosts": self.hosts}
527+
resource_config = {
528+
"current_host": host,
529+
"hosts": self.hosts,
530+
"network_interface_name": "eth0",
531+
"current_instance_type": self.instance_type,
532+
}
523533

524534
json_input_data_config = {}
525535
for c in input_data_config:

0 commit comments

Comments
 (0)