feat: add support for custom user image deployment

src/emd/cfn/sagemaker_realtime/template.yaml

@@ -34,6 +34,13 @@ Parameters:
     Type: Number
     Description: The target value for the autoscaling
     Default: 10
+  SageMakerEndpointName:
+    Type: String
+    Description: The name of the SageMaker Endpoint
+    Default: "noname"
+
+Conditions:
+  UseDefaultEndpointName: !Equals [!Ref SageMakerEndpointName, "noname"]
 
 Resources:
   ExecutionRole:
@@ -90,7 +97,10 @@ Resources:
   SageMakerEndpoint:
     Type: AWS::SageMaker::Endpoint
     Properties:
-      EndpointName: !Sub '${AWS::StackName}-endpoint'
+      EndpointName: !If 
+        - UseDefaultEndpointName
+        - !Sub '${AWS::StackName}-endpoint'
+        - !Ref SageMakerEndpointName
       EndpointConfigName: !GetAtt SageMakerEndpointConfig.EndpointConfigName
 
 

src/emd/commands/deploy.py

@@ -234,13 +234,34 @@ def deploy(
     ] = False,
     only_allow_local_deploy: Annotated[
         Optional[bool], typer.Option("--only-allow-local-deploy", help="only allow local instance")
-    ] = False
+    ] = False,
+    dockerfile_local_path: Annotated[
+        str, typer.Option("--dockerfile-local-path", help="Your custom Dockerfile path for building the model image, all files must be in the same directory")
+    ] = None,
 ):
     if only_allow_local_deploy:
         allow_local_deploy = True
         region = LOCAL_REGION
     else:
         region = get_current_region()
+
+    if dockerfile_local_path:
+        response = sdk_deploy(
+            model_id='custom-docker',
+            model_tag=f"{model_id}-{model_tag}",
+            instance_type=instance_type,
+            engine_type='custom',
+            framework_type='custom',
+            service_type='sagemaker_realtime',
+            region=region,
+            extra_params = extra_params,
+            env_stack_on_failure = "ROLLBACK",
+            force_env_stack_update = force_update_env_stack,
+            waiting_until_deploy_complete=True,
+            dockerfile_local_path=dockerfile_local_path,
+        )
+        return response
+
     vpc_id = None
     # ask model id
     model_id = ask_model_id(region,model_id=model_id)

src/emd/models/__init__.py

@@ -19,6 +19,7 @@
     asr,
     embeddings,
     reranks,
+    custom,
 )
 # text-2-image,text-2-video
 
@@ -27,3 +28,4 @@
 from . import services
 from . import model_series
 from . import frameworks
+

src/emd/models/model.py

@@ -74,7 +74,7 @@ def check_inf2_instance(cls,instance_type:str):
 class Engine(ModelBase):
     engine_type: EngineType
     engine_dockerfile_config: Union[dict,None] = Field(default_factory=dict)
-    engine_cls: str
+    engine_cls: Union[str,None] = None
     dockerfile_name: str = "Dockerfile"
     base_image_account_id: Union[str,None] = None
     base_image_host: Union[str,None] = None
@@ -181,15 +181,15 @@ class Model(ModelBase,Generic[T]):
     require_huggingface_token: bool = False
     modelscope_model_id: str = ""
     require_modelscope_token: bool = False
-    application_scenario: str
+    application_scenario: str = ""
     description: str =  ""
     model_type: ModelType = ModelType.LLM
     need_prepare_model: bool = True
     # download model files directly from s3
     model_files_s3_path: Union[str,None] = None
     model_files_local_path: Union[str,None] = None
     model_files_download_source: ModelFilesDownloadSource = ModelFilesDownloadSource.AUTO
-    model_series: ModelSeries
+    model_series: ModelSeries = None
     executable_config: Union[ExecutableConfig,None] = None
 
     @classmethod

src/emd/models/services.py

@@ -16,7 +16,8 @@
         "EngineType":"engine_type",
         "Region":"region",
         "MaxCapacity": ValueWithDefault(name="max_capacity",default=1),
-        "AutoScalingTargetValue": ValueWithDefault(name="auto_scaling_target_value",default=10)
+        "AutoScalingTargetValue": ValueWithDefault(name="auto_scaling_target_value",default=10),
+        "SageMakerEndpointName": ValueWithDefault(name="sagemaker_endpoint_name",default="noname")
     },
     name = "Amazon SageMaker AI Real-time inference",
     service_type=ServiceType.SAGEMAKER,

src/emd/models/utils/constants.py

@@ -32,6 +32,7 @@ class EngineType(ConstantBase):
     LLAMA_CPP = "llama.cpp"
     TGI = "tgi"
     LMDEPLOY = 'lmdeploy'
+    CUSTOM = "custom"
     KTRANFORMERS = 'ktransformers'
 
     # @classmethod

src/emd/sdk/deploy.py

@@ -1,15 +1,18 @@
 import json
 import os
+import io
 import time
 from typing import Optional
 
 import boto3
+import zipfile
 import sys
 
 from emd.constants import (
     CODEPIPELINE_NAME,
     ENV_STACK_NAME,
     MODEL_DEFAULT_TAG,
+    MODEL_STACK_NAME_PREFIX,
     VERSION,
     LOCAL_REGION
 )
@@ -57,11 +60,15 @@ def prepare_deploy(
     service_type=None,
     instance_type=None,
     region=None,
+    dockerfile_local_path=None
 ):
-    model: Model = Model.get_model(model_id)
-    model_stack_name = model.get_model_stack_name_prefix(
-        model_id, model_tag=model_tag
-    )
+    if dockerfile_local_path:
+        model_stack_name = f"{MODEL_STACK_NAME_PREFIX}-{model_id}-{model_tag}"
+    else:
+        model: Model = Model.get_model(model_id)
+        model_stack_name = model.get_model_stack_name_prefix(
+            model_id, model_tag=model_tag
+        )
     # check if model_id is inprogress in pipeline execution
     if check_stack_exists(model_stack_name):
         raise RuntimeError(
@@ -113,6 +120,7 @@ def deploy(
     env_stack_on_failure="ROLLBACK",
     force_env_stack_update=False,
     waiting_until_deploy_complete=True,
+    dockerfile_local_path=None,
 ) -> dict:
     # Check if AWS environment is properly configured
     if service_type == ServiceType.SAGEMAKER_OLDER:
@@ -132,9 +140,13 @@ def deploy(
         service_type=service_type,
         instance_type=instance_type,
         region=region,
+        dockerfile_local_path=dockerfile_local_path
     )
     # logger.info("Checking AWS environment...")
-    extra_params = extra_params or {}
+    if isinstance(extra_params, str):
+        extra_params = json.loads(extra_params)
+    else:
+        extra_params = extra_params or {}
     if model_stack_name is None:
         # stack_name_suffix = random_suffix()
         model_stack_name = (
@@ -168,23 +180,38 @@ def deploy(
     pipeline_name = pipeline_resources[0]["PhysicalResourceId"]
     logger.info("AWS environment is properly configured.")
 
-    model = Model.get_model(model_id)
-
-    # check instance,service,engine
-    supported_instances = model.supported_instance_types
-    assert (
-        instance_type in supported_instances
-    ), f"Instance type {instance_type} is not supported for model {model_id}"
-
-    supported_engines = model.supported_engine_types
-    assert (
-        engine_type in supported_engines
-    ), f"Engine type {engine_type} is not supported for model {model_id}"
-
-    supported_services = model.supported_service_types
-    assert (
-        service_type in supported_services
-    ), f"Service type {service_type} is not supported for model {model_id}"
+    if dockerfile_local_path:
+        if not os.path.exists(dockerfile_local_path):
+            raise FileNotFoundError(f"Dockerfile path {dockerfile_local_path} does not exist.")
+
+        # Create a zip file of the dockerfile directory
+        zip_buffer = zipped_dockerfile(dockerfile_local_path)
+
+        # Upload the zip file to S3
+        s3 = boto3.client('s3', region_name=region)
+        s3_key = f"emd_models/{model_id}-{model_tag}.zip"
+        s3.upload_fileobj(zip_buffer, bucket_name, s3_key)
+        extra_params["model_params"] = extra_params.get("model_params", {})
+        extra_params["model_params"]["custom_dockerfile_path"] = f"s3://{bucket_name}/{s3_key}"
+        logger.info(f"extra_params: {extra_params}")
+    else:
+        model = Model.get_model(model_id)
+
+        # check instance,service,engine
+        supported_instances = model.supported_instance_types
+        assert (
+            instance_type in supported_instances
+        ), f"Instance type {instance_type} is not supported for model {model_id}"
+
+        supported_engines = model.supported_engine_types
+        assert (
+            engine_type in supported_engines
+        ), f"Engine type {engine_type} is not supported for model {model_id}"
+
+        supported_services = model.supported_service_types
+        assert (
+            service_type in supported_services
+        ), f"Service type {service_type} is not supported for model {model_id}"
 
     # Start pipeline execution
     codepipeline = boto3.client("codepipeline", region_name=region)
@@ -329,3 +356,15 @@ def deploy_local(
     assert (
         os.system(pipeline_cmd) == 0
     ), f"run pipeline cmd failed: {pipeline_cmd}"
+
+def zipped_dockerfile(dockerfile_local_path):
+    zip_buffer = io.BytesIO()
+    with zipfile.ZipFile(zip_buffer, 'a', zipfile.ZIP_DEFLATED) as zipf:
+        dockerfile_dir = os.path.dirname(dockerfile_local_path)
+        for root, dirs, files in os.walk(dockerfile_dir):
+            for file in files:
+                file_path = os.path.join(root, file)
+                arcname = os.path.relpath(file_path, dockerfile_dir)
+                zipf.write(file_path, arcname)
+    zip_buffer.seek(0)
+    return zip_buffer