When using the model_id with the DJLModel class in the SageMaker Python SDK, the entry_point is not working, and the code specified is not loaded into the container

[itaymelamed]

Describe the bug
When using the microsoft/Phi-3-small-8k-instruct with the DJLModel class in the SageMaker Python SDK, the entry_point is not working, and the code specified is not loaded into the container. When endpoint is created it fails since in can't locate /opt/ml/model/code/model.py.

To reproduce
Steps to reproduce the behavior:

1. Set up a SageMaker environment with the required IAM roles and permissions.

2. Use the following code snippet to create a DJLModel:

   from sagemaker.djl import DJLModel
   
   djl_model = DJLModel(
       model_id="microsoft/Phi-3-small-128k-instruct/",
       sagemaker_session=sagemaker_session,
       djl_version="0.29.0",
       role=iam_role,
       task="text-generation",
       engine="Python",
       entry_point="model.py",
       source_dir="code",
       env={
           "OPTION_ROLLING_BATCH": "vllm",
           "TENSOR_PARALLEL_DEGREE": "1",
           "OPTION_MAX_ROLLING_BATCH_SIZE": "2",
           "OPTION_DTYPE": "fp16",
           "HF_MODEL_TRUST_REMOTE_CODE": "True",
           "DJL_ENTRY_POINT": "/opt/ml/model/code/model.py"
       }
   )

3. Deploy the model using SageMaker.

Expected behavior
The entry_point specified should be correctly loaded into the container, allowing the model to function as intended.

System information

• SageMaker Python SDK version:2.229.0
• Framework name (e.g., PyTorch) or algorithm (e.g., KMeans): DJL
• Framework version: 0.29.0
• Python version: 3.11
• CPU or GPU: GPU
• Custom Docker image (Y/N): N