Skip to content

Commit 2bd30c4

Browse files
CaptainiaCaptainia
authored
feat: add new images for HF TGI release (#4677)
* chore: add new images for HF TGI release * test
1 parent 2f2be05 commit 2bd30c4

File tree

5 files changed

+168
-5
lines changed

5 files changed

+168
-5
lines changed

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

Lines changed: 30 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -4,7 +4,7 @@
44
"inf2"
55
],
66
"version_aliases": {
7-
"0.0": "0.0.16"
7+
"0.0": "0.0.22"
88
},
99
"versions": {
1010
"0.0.16": {
@@ -180,6 +180,35 @@
180180
"container_version": {
181181
"inf2": "ubuntu22.04"
182182
}
183+
},
184+
"0.0.22": {
185+
"py_versions": [
186+
"py310"
187+
],
188+
"registries": {
189+
"ap-northeast-1": "763104351884",
190+
"ap-south-1": "763104351884",
191+
"ap-south-2": "772153158452",
192+
"ap-southeast-1": "763104351884",
193+
"ap-southeast-2": "763104351884",
194+
"ap-southeast-4": "457447274322",
195+
"eu-central-1": "763104351884",
196+
"eu-central-2": "380420809688",
197+
"eu-south-2": "503227376785",
198+
"eu-west-1": "763104351884",
199+
"eu-west-3": "763104351884",
200+
"il-central-1": "780543022126",
201+
"sa-east-1": "763104351884",
202+
"us-east-1": "763104351884",
203+
"us-east-2": "763104351884",
204+
"us-west-2": "763104351884",
205+
"ca-west-1": "204538143572"
206+
},
207+
"tag_prefix": "2.1.2-optimum0.0.22",
208+
"repository": "huggingface-pytorch-tgi-inference",
209+
"container_version": {
210+
"inf2": "ubuntu22.04"
211+
}
183212
}
184213
}
185214
}

‎src/sagemaker/image_uri_config/huggingface-llm.json‎

Lines changed: 48 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -12,7 +12,7 @@
1212
"1.2": "1.2.0",
1313
"1.3": "1.3.3",
1414
"1.4": "1.4.5",
15-
"2.0": "2.0.1"
15+
"2.0": "2.0.2"
1616
},
1717
"versions": {
1818
"0.6.0": {
@@ -625,6 +625,53 @@
625625
"container_version": {
626626
"gpu": "cu121-ubuntu22.04"
627627
}
628+
},
629+
"2.0.2": {
630+
"py_versions": [
631+
"py310"
632+
],
633+
"registries": {
634+
"af-south-1": "626614931356",
635+
"il-central-1": "780543022126",
636+
"ap-east-1": "871362719292",
637+
"ap-northeast-1": "763104351884",
638+
"ap-northeast-2": "763104351884",
639+
"ap-northeast-3": "364406365360",
640+
"ap-south-1": "763104351884",
641+
"ap-south-2": "772153158452",
642+
"ap-southeast-1": "763104351884",
643+
"ap-southeast-2": "763104351884",
644+
"ap-southeast-3": "907027046896",
645+
"ap-southeast-4": "457447274322",
646+
"ca-central-1": "763104351884",
647+
"cn-north-1": "727897471807",
648+
"cn-northwest-1": "727897471807",
649+
"eu-central-1": "763104351884",
650+
"eu-central-2": "380420809688",
651+
"eu-north-1": "763104351884",
652+
"eu-west-1": "763104351884",
653+
"eu-west-2": "763104351884",
654+
"eu-west-3": "763104351884",
655+
"eu-south-1": "692866216735",
656+
"eu-south-2": "503227376785",
657+
"me-south-1": "217643126080",
658+
"me-central-1": "914824155844",
659+
"sa-east-1": "763104351884",
660+
"us-east-1": "763104351884",
661+
"us-east-2": "763104351884",
662+
"us-gov-east-1": "446045086412",
663+
"us-gov-west-1": "442386744353",
664+
"us-iso-east-1": "886529160074",
665+
"us-isob-east-1": "094389454867",
666+
"us-west-1": "763104351884",
667+
"us-west-2": "763104351884",
668+
"ca-west-1": "204538143572"
669+
},
670+
"tag_prefix": "2.3.0-tgi2.0.2",
671+
"repository": "huggingface-pytorch-tgi-inference",
672+
"container_version": {
673+
"gpu": "cu121-ubuntu22.04"
674+
}
628675
}
629676
}
630677
}

‎src/sagemaker/image_uri_config/huggingface-neuronx.json‎

Lines changed: 87 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -5,7 +5,8 @@
55
],
66
"version_aliases": {
77
"4.28": "4.28.1",
8-
"4.34": "4.34.1"
8+
"4.34": "4.34.1",
9+
"4.36": "4.36.2"
910
},
1011
"versions": {
1112
"4.28.1": {
@@ -79,6 +80,42 @@
7980
"sdk2.15.0"
8081
]
8182
}
83+
},
84+
"4.36.2": {
85+
"version_aliases": {
86+
"pytorch1.13": "pytorch1.13.1"
87+
},
88+
"pytorch1.13.1": {
89+
"py_versions": [
90+
"py310"
91+
],
92+
"repository": "huggingface-pytorch-inference-neuronx",
93+
"registries": {
94+
"ap-northeast-1": "763104351884",
95+
"ap-south-1": "763104351884",
96+
"ap-south-2": "772153158452",
97+
"ap-southeast-1": "763104351884",
98+
"ap-southeast-2": "763104351884",
99+
"ap-southeast-4": "457447274322",
100+
"eu-central-1": "763104351884",
101+
"eu-central-2": "380420809688",
102+
"eu-south-2": "503227376785",
103+
"eu-west-1": "763104351884",
104+
"eu-west-3": "763104351884",
105+
"il-central-1": "780543022126",
106+
"sa-east-1": "763104351884",
107+
"us-east-1": "763104351884",
108+
"us-east-2": "763104351884",
109+
"us-west-2": "763104351884",
110+
"ca-west-1": "204538143572"
111+
},
112+
"container_version": {
113+
"inf": "ubuntu20.04"
114+
},
115+
"sdk_versions": [
116+
"sdk2.18.0"
117+
]
118+
}
82119
}
83120
}
84121
},
@@ -198,7 +235,8 @@
198235
},
199236
"4.36.2": {
200237
"version_aliases": {
201-
"pytorch1.13": "pytorch1.13.1"
238+
"pytorch1.13": "pytorch1.13.1",
239+
"pytorch2.1": "pytorch2.1.2"
202240
},
203241
"pytorch1.13.1": {
204242
"py_versions": [
@@ -246,6 +284,53 @@
246284
"sdk_versions": [
247285
"sdk2.16.1"
248286
]
287+
},
288+
"pytorch2.1.2": {
289+
"py_versions": [
290+
"py310"
291+
],
292+
"repository": "huggingface-pytorch-inference-neuronx",
293+
"registries": {
294+
"af-south-1": "626614931356",
295+
"il-central-1": "780543022126",
296+
"ap-east-1": "871362719292",
297+
"ap-northeast-1": "763104351884",
298+
"ap-northeast-2": "763104351884",
299+
"ap-northeast-3": "364406365360",
300+
"ap-south-1": "763104351884",
301+
"ap-south-2": "772153158452",
302+
"ap-southeast-1": "763104351884",
303+
"ap-southeast-2": "763104351884",
304+
"ap-southeast-4": "457447274322",
305+
"ca-central-1": "763104351884",
306+
"cn-north-1": "727897471807",
307+
"cn-northwest-1": "727897471807",
308+
"eu-central-1": "763104351884",
309+
"eu-central-2": "380420809688",
310+
"eu-north-1": "763104351884",
311+
"eu-west-1": "763104351884",
312+
"eu-west-2": "763104351884",
313+
"eu-west-3": "763104351884",
314+
"eu-south-1": "692866216735",
315+
"eu-south-2": "503227376785",
316+
"me-south-1": "217643126080",
317+
"sa-east-1": "763104351884",
318+
"us-east-1": "763104351884",
319+
"us-east-2": "763104351884",
320+
"us-gov-east-1": "446045086412",
321+
"us-gov-west-1": "442386744353",
322+
"us-iso-east-1": "886529160074",
323+
"us-isob-east-1": "094389454867",
324+
"us-west-1": "763104351884",
325+
"us-west-2": "763104351884",
326+
"ca-west-1": "204538143572"
327+
},
328+
"container_version": {
329+
"inf": "ubuntu20.04"
330+
},
331+
"sdk_versions": [
332+
"sdk2.18.0"
333+
]
249334
}
250335
}
251336
}

‎src/sagemaker/jumpstart/utils.py‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -123,7 +123,7 @@ def get_jumpstart_gated_content_bucket(
123123
def get_jumpstart_content_bucket(
124124
region: str = constants.JUMPSTART_DEFAULT_REGION_NAME,
125125
) -> str:
126-
"""Returns regionalized content bucket name for JumpStart.
126+
"""Returns the regionalized content bucket name for JumpStart.
127127
128128
Raises:
129129
ValueError: If JumpStart is not launched in ``region``.

‎tests/unit/sagemaker/image_uris/test_huggingface_llm.py‎

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -33,6 +33,7 @@
3333
"1.4.5": "2.1.1-tgi1.4.5-gpu-py310-cu121-ubuntu22.04",
3434
"2.0.0": "2.1.1-tgi2.0.0-gpu-py310-cu121-ubuntu22.04",
3535
"2.0.1": "2.1.1-tgi2.0.1-gpu-py310-cu121-ubuntu22.04",
36+
"2.0.2": "2.3.0-tgi2.0.2-gpu-py310-cu121-ubuntu22.04",
3637
},
3738
"inf2": {
3839
"0.0.16": "1.13.1-optimum0.0.16-neuronx-py310-ubuntu22.04",
@@ -41,6 +42,7 @@
4142
"0.0.19": "1.13.1-optimum0.0.19-neuronx-py310-ubuntu22.04",
4243
"0.0.20": "1.13.1-optimum0.0.20-neuronx-py310-ubuntu22.04",
4344
"0.0.21": "1.13.1-optimum0.0.21-neuronx-py310-ubuntu22.04",
45+
"0.0.22": "2.1.2-optimum0.0.22-neuronx-py310-ubuntu22.04",
4446
},
4547
}
4648

0 commit comments

Comments
 (0)