diff --git a/src/sagemaker/djl_inference/model.py b/src/sagemaker/djl_inference/model.py index 3fa523c605..8c724a6502 100644 --- a/src/sagemaker/djl_inference/model.py +++ b/src/sagemaker/djl_inference/model.py @@ -43,7 +43,7 @@ def __init__( self, model_id: Optional[str] = None, engine: Optional[str] = None, - djl_version: str = "0.28.0", + djl_version: str = "latest", djl_framework: Optional[str] = None, task: Optional[str] = None, dtype: Optional[str] = None, diff --git a/src/sagemaker/image_uri_config/djl-lmi.json b/src/sagemaker/image_uri_config/djl-lmi.json index 325327ca8e..c55ff35bed 100644 --- a/src/sagemaker/image_uri_config/djl-lmi.json +++ b/src/sagemaker/image_uri_config/djl-lmi.json @@ -2,6 +2,9 @@ "scope": [ "inference" ], + "version_aliases": { + "latest": "0.29.0" + }, "versions": { "0.29.0": { "registries": { diff --git a/src/sagemaker/image_uri_config/djl-neuronx.json b/src/sagemaker/image_uri_config/djl-neuronx.json index 8a59426ca9..d62a4b0b1e 100644 --- a/src/sagemaker/image_uri_config/djl-neuronx.json +++ b/src/sagemaker/image_uri_config/djl-neuronx.json @@ -2,6 +2,9 @@ "scope": [ "inference" ], + "version_aliases": { + "latest": "0.29.0" + }, "versions": { "0.29.0": { "registries": { diff --git a/src/sagemaker/image_uri_config/djl-tensorrtllm.json b/src/sagemaker/image_uri_config/djl-tensorrtllm.json index 68db982632..47088febfc 100644 --- a/src/sagemaker/image_uri_config/djl-tensorrtllm.json +++ b/src/sagemaker/image_uri_config/djl-tensorrtllm.json @@ -2,6 +2,9 @@ "scope": [ "inference" ], + "version_aliases": { + "latest": "0.29.0" + }, "versions": { "0.29.0": { "registries": { diff --git a/tests/unit/sagemaker/image_uris/test_djl.py b/tests/unit/sagemaker/image_uris/test_djl.py index 6457fe044f..887b575fdf 100644 --- a/tests/unit/sagemaker/image_uris/test_djl.py +++ b/tests/unit/sagemaker/image_uris/test_djl.py @@ -18,12 +18,7 @@ @pytest.mark.parametrize( "load_config_and_file_name", - [ - "djl-neuronx.json", - "djl-fastertransformer.json", - "djl-deepspeed.json", - "djl-tensorrtllm.json", - ], + ["djl-neuronx.json", "djl-tensorrtllm.json", "djl-lmi.json"], indirect=True, ) def test_djl_uris(load_config_and_file_name): diff --git a/tests/unit/test_djl_inference.py b/tests/unit/test_djl_inference.py index 6b0f5a6f92..abdd86fbfc 100644 --- a/tests/unit/test_djl_inference.py +++ b/tests/unit/test_djl_inference.py @@ -26,7 +26,7 @@ HF_MODEL_ID = "hf_hub_model_id" ROLE = "dummy_role" REGION = "us-west-2" -VERSION = "0.28.0" +VERSION = "latest" LMI_IMAGE_URI = image_uris.retrieve(framework="djl-lmi", version=VERSION, region=REGION) TRT_IMAGE_URI = image_uris.retrieve(framework="djl-tensorrtllm", version=VERSION, region=REGION)