Merge branch 'master' into master

Author: mwfongAWS

src/sagemaker/djl_inference/model.py

@@ -781,7 +781,7 @@ def serving_image_uri(self, region_name):
             str: The appropriate image URI based on the given parameters.
         """
         if not self.djl_version:
-            self.djl_version = "0.23.0"
+            self.djl_version = "0.24.0"
 
         return image_uris.retrieve(
             self._framework(),

src/sagemaker/huggingface/llm_utils.py

@@ -51,6 +51,6 @@ def get_huggingface_llm_image_uri(
             image_scope="inference",
         )
     if backend == "lmi":
-        version = version or "0.23.0"
+        version = version or "0.24.0"
         return image_uris.retrieve(framework="djl-deepspeed", region=region, version=version)
     raise ValueError("Unsupported backend: %s" % backend)

src/sagemaker/image_uri_config/djl-deepspeed.json

@@ -1,6 +1,37 @@
 {
     "scope": ["inference"],
     "versions": {
+         "0.24.0": {
+            "registries": {
+                "af-south-1": "626614931356",
+                "il-central-1": "780543022126",
+                "ap-east-1": "871362719292",
+                "ap-northeast-1": "763104351884",
+                "ap-northeast-2": "763104351884",
+                "ap-northeast-3": "364406365360",
+                "ap-south-1": "763104351884",
+                "ap-southeast-1": "763104351884",
+                "ap-southeast-2": "763104351884",
+                "ap-southeast-3": "907027046896",
+                "ca-central-1": "763104351884",
+                "cn-north-1": "727897471807",
+                "cn-northwest-1": "727897471807",
+                "eu-central-1": "763104351884",
+                "eu-north-1": "763104351884",
+                "eu-west-1": "763104351884",
+                "eu-west-2": "763104351884",
+                "eu-west-3": "763104351884",
+                "eu-south-1": "692866216735",
+                "me-south-1": "217643126080",
+                "sa-east-1": "763104351884",
+                "us-east-1": "763104351884",
+                "us-east-2": "763104351884",
+                "us-west-1": "763104351884",
+                "us-west-2": "763104351884"
+            },
+            "repository": "djl-inference",
+            "tag_prefix": "0.24.0-deepspeed0.10.0-cu118"
+        },
          "0.23.0": {
             "registries": {
                 "af-south-1": "626614931356",

src/sagemaker/image_uri_config/djl-fastertransformer.json

@@ -1,6 +1,37 @@
 {
     "scope": ["inference"],
     "versions": {
+        "0.24.0": {
+            "registries": {
+                "af-south-1": "626614931356",
+                "il-central-1": "780543022126",
+                "ap-east-1": "871362719292",
+                "ap-northeast-1": "763104351884",
+                "ap-northeast-2": "763104351884",
+                "ap-northeast-3": "364406365360",
+                "ap-south-1": "763104351884",
+                "ap-southeast-1": "763104351884",
+                "ap-southeast-2": "763104351884",
+                "ap-southeast-3": "907027046896",
+                "ca-central-1": "763104351884",
+                "cn-north-1": "727897471807",
+                "cn-northwest-1": "727897471807",
+                "eu-central-1": "763104351884",
+                "eu-north-1": "763104351884",
+                "eu-west-1": "763104351884",
+                "eu-west-2": "763104351884",
+                "eu-west-3": "763104351884",
+                "eu-south-1": "692866216735",
+                "me-south-1": "217643126080",
+                "sa-east-1": "763104351884",
+                "us-east-1": "763104351884",
+                "us-east-2": "763104351884",
+                "us-west-1": "763104351884",
+                "us-west-2": "763104351884"
+            },
+            "repository": "djl-inference",
+            "tag_prefix": "0.24.0-fastertransformer5.3.0-cu118"
+        },
         "0.23.0": {
             "registries": {
                 "af-south-1": "626614931356",

tests/unit/sagemaker/image_uris/test_djl.py

@@ -42,9 +42,8 @@
     "us-west-1": "763104351884",
     "us-west-2": "763104351884",
 }
-DJL_DEEPSPEED_VERSIONS = ["0.23.0", "0.22.1", "0.21.0", "0.20.0", "0.19.0"]
-DJL_FASTERTRANSFORMER_VERSIONS = ["0.23.0", "0.22.1", "0.21.0"]
-DJL_NEURONX_VERSIONS = ["0.23.0", "0.22.1"]
+DJL_DEEPSPEED_VERSIONS = ["0.24.0", "0.23.0", "0.22.1", "0.21.0", "0.20.0", "0.19.0"]
+DJL_FASTERTRANSFORMER_VERSIONS = ["0.24.0", "0.23.0", "0.22.1", "0.21.0"]
 DJL_NEURONX_VERSIONS = ["0.24.0", "0.23.0", "0.22.1"]
 DJL_VERSIONS_TO_FRAMEWORK = {
     "0.19.0": {"djl-deepspeed": "deepspeed0.7.3-cu113"},
@@ -64,6 +63,8 @@
         "djl-neuronx": "neuronx-sdk2.12.0",
     },
     "0.24.0": {
+        "djl-deepspeed": "deepspeed0.10.0-cu118",
+        "djl-fastertransformer": "fastertransformer5.3.0-cu118",
         "djl-neuronx": "neuronx-sdk2.14.1",
     },
 }

tests/unit/sagemaker/image_uris/test_huggingface_llm.py

@@ -45,15 +45,15 @@
     "us-west-2": "763104351884",
 }
 HF_VERSIONS = ["0.6.0", "0.8.2", "0.9.3", "1.0.3", "1.1.0"]
-LMI_VERSIONS = ["0.23.0"]
+LMI_VERSIONS = ["0.24.0"]
 HF_VERSIONS_MAPPING = {
     "0.6.0": "2.0.0-tgi0.6.0-gpu-py39-cu118-ubuntu20.04",
     "0.8.2": "2.0.0-tgi0.8.2-gpu-py39-cu118-ubuntu20.04",
     "0.9.3": "2.0.1-tgi0.9.3-gpu-py39-cu118-ubuntu20.04",
     "1.0.3": "2.0.1-tgi1.0.3-gpu-py39-cu118-ubuntu20.04",
     "1.1.0": "2.0.1-tgi1.1.0-gpu-py39-cu118-ubuntu20.04",
 }
-LMI_VERSIONS_MAPPING = {"0.23.0": "deepspeed0.9.5-cu118"}
+LMI_VERSIONS_MAPPING = {"0.24.0": "deepspeed0.10.0-cu118"}
 
 
 @pytest.mark.parametrize("version", HF_VERSIONS)

tests/unit/test_djl_inference.py

@@ -46,7 +46,7 @@
 ROLE = "dummy_role"
 REGION = "us-west-2"
 BUCKET = "mybucket"
-IMAGE_URI = "763104351884.dkr.ecr.us-west-2.amazon.com/djl-inference:0.23.0-deepspeed0.9.5-cu118"
+IMAGE_URI = "763104351884.dkr.ecr.us-west-2.amazon.com/djl-inference:0.24.0-deepspeed0.10.0-cu118"
 GPU_INSTANCE = "ml.g5.12xlarge"