Skip to content

Commit 4948edd

Browse files
pagezyhfzhaoqizqwang
pagezyhf
and
zhaoqizqwang
authored
add HF Optimum Neuron DLCs (#5309)
* add image * inf on dlc * neuron tgi dlcs * fix test --------- Co-authored-by: Zhaoqi <[email protected]>
1 parent 43fcb4e commit 4948edd

File tree

3 files changed

+208
-7
lines changed

3 files changed

+208
-7
lines changed

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

Lines changed: 111 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -4,7 +4,9 @@
44
"inf2"
55
],
66
"version_aliases": {
7-
"0.0": "0.0.28"
7+
"0.0": "0.0.28",
8+
"0.2": "0.2.0",
9+
"0.3": "0.3.0"
810
},
911
"versions": {
1012
"0.0.16": {
@@ -654,6 +656,114 @@
654656
"container_version": {
655657
"inf2": "ubuntu22.04"
656658
}
659+
},
660+
"0.2.0": {
661+
"py_versions": [
662+
"py310"
663+
],
664+
"registries": {
665+
"af-south-1": "626614931356",
666+
"ap-east-1": "871362719292",
667+
"ap-east-2": "975050140332",
668+
"ap-northeast-1": "763104351884",
669+
"ap-northeast-2": "763104351884",
670+
"ap-northeast-3": "364406365360",
671+
"ap-south-1": "763104351884",
672+
"ap-south-2": "772153158452",
673+
"ap-southeast-1": "763104351884",
674+
"ap-southeast-2": "763104351884",
675+
"ap-southeast-3": "907027046896",
676+
"ap-southeast-4": "457447274322",
677+
"ap-southeast-5": "550225433462",
678+
"ap-southeast-6": "633930458069",
679+
"ap-southeast-7": "590183813437",
680+
"ca-central-1": "763104351884",
681+
"ca-west-1": "204538143572",
682+
"cn-north-1": "727897471807",
683+
"cn-northwest-1": "727897471807",
684+
"eu-central-1": "763104351884",
685+
"eu-central-2": "380420809688",
686+
"eu-north-1": "763104351884",
687+
"eu-south-1": "692866216735",
688+
"eu-south-2": "503227376785",
689+
"eu-west-1": "763104351884",
690+
"eu-west-2": "763104351884",
691+
"eu-west-3": "763104351884",
692+
"il-central-1": "780543022126",
693+
"me-central-1": "914824155844",
694+
"me-south-1": "217643126080",
695+
"mx-central-1": "637423239942",
696+
"sa-east-1": "763104351884",
697+
"us-east-1": "763104351884",
698+
"us-east-2": "763104351884",
699+
"us-gov-east-1": "446045086412",
700+
"us-gov-west-1": "442386744353",
701+
"us-iso-east-1": "886529160074",
702+
"us-isob-east-1": "094389454867",
703+
"us-isof-east-1": "303241398832",
704+
"us-isof-south-1": "454834333376",
705+
"us-west-1": "763104351884",
706+
"us-west-2": "763104351884"
707+
},
708+
"tag_prefix": "2.5.1-optimum3.3.4",
709+
"repository": "huggingface-pytorch-tgi-inference",
710+
"container_version": {
711+
"inf2": "ubuntu22.04"
712+
}
713+
},
714+
"0.3.0": {
715+
"py_versions": [
716+
"py310"
717+
],
718+
"registries": {
719+
"af-south-1": "626614931356",
720+
"ap-east-1": "871362719292",
721+
"ap-east-2": "975050140332",
722+
"ap-northeast-1": "763104351884",
723+
"ap-northeast-2": "763104351884",
724+
"ap-northeast-3": "364406365360",
725+
"ap-south-1": "763104351884",
726+
"ap-south-2": "772153158452",
727+
"ap-southeast-1": "763104351884",
728+
"ap-southeast-2": "763104351884",
729+
"ap-southeast-3": "907027046896",
730+
"ap-southeast-4": "457447274322",
731+
"ap-southeast-5": "550225433462",
732+
"ap-southeast-6": "633930458069",
733+
"ap-southeast-7": "590183813437",
734+
"ca-central-1": "763104351884",
735+
"ca-west-1": "204538143572",
736+
"cn-north-1": "727897471807",
737+
"cn-northwest-1": "727897471807",
738+
"eu-central-1": "763104351884",
739+
"eu-central-2": "380420809688",
740+
"eu-north-1": "763104351884",
741+
"eu-south-1": "692866216735",
742+
"eu-south-2": "503227376785",
743+
"eu-west-1": "763104351884",
744+
"eu-west-2": "763104351884",
745+
"eu-west-3": "763104351884",
746+
"il-central-1": "780543022126",
747+
"me-central-1": "914824155844",
748+
"me-south-1": "217643126080",
749+
"mx-central-1": "637423239942",
750+
"sa-east-1": "763104351884",
751+
"us-east-1": "763104351884",
752+
"us-east-2": "763104351884",
753+
"us-gov-east-1": "446045086412",
754+
"us-gov-west-1": "442386744353",
755+
"us-iso-east-1": "886529160074",
756+
"us-isob-east-1": "094389454867",
757+
"us-isof-east-1": "303241398832",
758+
"us-isof-south-1": "454834333376",
759+
"us-west-1": "763104351884",
760+
"us-west-2": "763104351884"
761+
},
762+
"tag_prefix": "2.7.0-optimum3.3.6",
763+
"repository": "huggingface-pytorch-tgi-inference",
764+
"container_version": {
765+
"inf2": "ubuntu22.04"
766+
}
657767
}
658768
}
659769
}

‎src/sagemaker/image_uri_config/huggingface-neuronx.json‎

Lines changed: 94 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -8,7 +8,8 @@
88
"4.34": "4.34.1",
99
"4.36": "4.36.2",
1010
"4.43": "4.43.2",
11-
"4.48": "4.48.1"
11+
"4.48": "4.48.1",
12+
"4.51": "4.51.0"
1213
},
1314
"versions": {
1415
"4.28.1": {
@@ -63,7 +64,7 @@
6364
"py_versions": [
6465
"py310"
6566
],
66-
"repository": "huggingface-pytorch-inference-neuronx",
67+
"repository": "huggingface-pytorch-training-neuronx",
6768
"registries": {
6869
"ap-east-2": "975050140332",
6970
"ap-northeast-1": "763104351884",
@@ -107,7 +108,7 @@
107108
"py_versions": [
108109
"py310"
109110
],
110-
"repository": "huggingface-pytorch-inference-neuronx",
111+
"repository": "huggingface-pytorch-training-neuronx",
111112
"registries": {
112113
"ap-east-2": "975050140332",
113114
"ap-northeast-1": "763104351884",
@@ -151,7 +152,7 @@
151152
"py_versions": [
152153
"py310"
153154
],
154-
"repository": "huggingface-pytorch-inference-neuronx",
155+
"repository": "huggingface-pytorch-training-neuronx",
155156
"registries": {
156157
"ap-northeast-1": "763104351884",
157158
"ap-south-1": "763104351884",
@@ -194,7 +195,7 @@
194195
"py_versions": [
195196
"py310"
196197
],
197-
"repository": "huggingface-pytorch-inference-neuronx",
198+
"repository": "huggingface-pytorch-training-neuronx",
198199
"registries": {
199200
"ap-northeast-1": "763104351884",
200201
"ap-south-1": "763104351884",
@@ -228,6 +229,49 @@
228229
"sdk2.20.0"
229230
]
230231
}
232+
},
233+
"4.51.0": {
234+
"version_aliases": {
235+
"pytorch2.7": "pytorch2.7.0"
236+
},
237+
"pytorch2.7.0": {
238+
"py_versions": [
239+
"py310"
240+
],
241+
"repository": "huggingface-pytorch-training-neuronx",
242+
"registries": {
243+
"ap-northeast-1": "763104351884",
244+
"ap-south-1": "763104351884",
245+
"ap-south-2": "772153158452",
246+
"ap-southeast-1": "763104351884",
247+
"ap-southeast-2": "763104351884",
248+
"ap-southeast-4": "457447274322",
249+
"ap-southeast-5": "550225433462",
250+
"ap-southeast-7": "590183813437",
251+
"cn-north-1": "727897471807",
252+
"cn-northwest-1": "727897471807",
253+
"eu-central-1": "763104351884",
254+
"eu-central-2": "380420809688",
255+
"eu-south-2": "503227376785",
256+
"eu-west-1": "763104351884",
257+
"eu-west-3": "763104351884",
258+
"il-central-1": "780543022126",
259+
"mx-central-1":"637423239942",
260+
"sa-east-1": "763104351884",
261+
"us-east-1": "763104351884",
262+
"us-east-2": "763104351884",
263+
"us-gov-east-1": "446045086412",
264+
"us-gov-west-1": "442386744353",
265+
"us-west-2": "763104351884",
266+
"ca-west-1": "204538143572"
267+
},
268+
"container_version": {
269+
"inf": "ubuntu22.04"
270+
},
271+
"sdk_versions": [
272+
"sdk2.24.1"
273+
]
274+
}
231275
}
232276
}
233277
},
@@ -239,7 +283,8 @@
239283
"4.28": "4.28.1",
240284
"4.34": "4.34.1",
241285
"4.36": "4.36.2",
242-
"4.43": "4.43.2"
286+
"4.43": "4.43.2",
287+
"4.51": "4.51.3"
243288
},
244289
"versions": {
245290
"4.28.1": {
@@ -504,6 +549,49 @@
504549
"sdk2.20.0"
505550
]
506551
}
552+
},
553+
"4.51.3": {
554+
"version_aliases": {
555+
"pytorch2.7": "pytorch2.7.1"
556+
},
557+
"pytorch2.7.1": {
558+
"py_versions": [
559+
"py310"
560+
],
561+
"repository": "huggingface-pytorch-inference-neuronx",
562+
"registries": {
563+
"ap-northeast-1": "763104351884",
564+
"ap-south-1": "763104351884",
565+
"ap-south-2": "772153158452",
566+
"ap-southeast-1": "763104351884",
567+
"ap-southeast-2": "763104351884",
568+
"ap-southeast-4": "457447274322",
569+
"ap-southeast-5": "550225433462",
570+
"ap-southeast-7": "590183813437",
571+
"cn-north-1": "727897471807",
572+
"cn-northwest-1": "727897471807",
573+
"eu-central-1": "763104351884",
574+
"eu-central-2": "380420809688",
575+
"eu-south-2": "503227376785",
576+
"eu-west-1": "763104351884",
577+
"eu-west-3": "763104351884",
578+
"il-central-1": "780543022126",
579+
"mx-central-1":"637423239942",
580+
"sa-east-1": "763104351884",
581+
"us-east-1": "763104351884",
582+
"us-east-2": "763104351884",
583+
"us-gov-east-1": "446045086412",
584+
"us-gov-west-1": "442386744353",
585+
"us-west-2": "763104351884",
586+
"ca-west-1": "204538143572"
587+
},
588+
"container_version": {
589+
"inf": "ubuntu22.04"
590+
},
591+
"sdk_versions": [
592+
"sdk2.24.1"
593+
]
594+
}
507595
}
508596
}
509597
}

‎tests/unit/sagemaker/image_uris/test_huggingface_llm.py‎

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -75,6 +75,9 @@
7575
"0.0.24": "2.1.2-optimum0.0.24-neuronx-py310-ubuntu22.04",
7676
"0.0.25": "2.1.2-optimum0.0.25-neuronx-py310-ubuntu22.04",
7777
"0.0.27": "2.1.2-optimum0.0.27-neuronx-py310-ubuntu22.04",
78+
"0.0.28": "2.1.2-optimum0.0.28-neuronx-py310-ubuntu22.04",
79+
"0.2.0": "2.5.1-optimum3.3.4-neuronx-py310-ubuntu22.04",
80+
"0.3.0": "2.7.0-optimum3.3.6-neuronx-py310-ubuntu22.04",
7881
},
7982
}
8083

0 commit comments

Comments
 (0)