Update

amzn-choeric · amzn-choeric · commit ad3992fd17c4 · 2024-01-17T14:57:40.000-05:00
diff --git a/src/sagemaker/image_uri_config/huggingface-llm-neuronx.json b/src/sagemaker/image_uri_config/huggingface-llm-neuronx.json
@@ -1,7 +1,7 @@
 {
     "inference": {
         "processors": [
-            "inf"
+            "inf2"
         ],
         "version_aliases": {
             "0.0": "0.0.16"
@@ -30,10 +30,10 @@
                     "us-west-2": "763104351884",
                     "ca-west-1": "204538143572"
                 },
-                "tag_prefix": "2.1.1-optimum0.0.16",
+                "tag_prefix": "1.13.1-optimum0.0.16",
                 "repository": "huggingface-pytorch-tgi-inference",
                 "container_version": {
-                    "inf": "ubuntu22.04"
+                    "inf2": "ubuntu22.04"
                 }
             }
         }
diff --git a/tests/unit/sagemaker/image_uris/test_huggingface_llm.py b/tests/unit/sagemaker/image_uris/test_huggingface_llm.py
@@ -29,8 +29,8 @@
         "1.3.1": "2.1.1-tgi1.3.1-gpu-py310-cu121-ubuntu20.04",
         "1.3.3": "2.1.1-tgi1.3.3-gpu-py310-cu121-ubuntu20.04",
     },
-    "inf": {
-        "0.0.16": "2.1.1-optimum0.0.16-neuronx-py310-ubuntu22.04",
+    "inf2": {
+        "0.0.16": "1.13.1-optimum0.0.16-neuronx-py310-ubuntu22.04",
     },
 }
 

Original file line number	Diff line number	Diff line change
`@@ -1,7 +1,7 @@`
`1`	`1`	`{`
`2`	`2`	`"inference": {`
`3`	`3`	`"processors": [`
`4`		`- "inf"`
	`4`	`+ "inf2"`
`5`	`5`	`],`
`6`	`6`	`"version_aliases": {`
`7`	`7`	`"0.0": "0.0.16"`
`@@ -30,10 +30,10 @@`
`30`	`30`	`"us-west-2": "763104351884",`
`31`	`31`	`"ca-west-1": "204538143572"`
`32`	`32`	`},`
`33`		`- "tag_prefix": "2.1.1-optimum0.0.16",`
	`33`	`+ "tag_prefix": "1.13.1-optimum0.0.16",`
`34`	`34`	`"repository": "huggingface-pytorch-tgi-inference",`
`35`	`35`	`"container_version": {`
`36`		`- "inf": "ubuntu22.04"`
	`36`	`+ "inf2": "ubuntu22.04"`
`37`	`37`	`}`
`38`	`38`	`}`
`39`	`39`	`}`