我正在尝试使用sagemaker多容器选项来部署模型。我使用的是最新的sagemaker版本和PyTorch型号。从我的角度来看，这看起来像是版本不匹配。
create_model_response = sm_client.create_model(
ModelName="multi-container3",
Containers=[pytorch_container, pytorch_container2],
InferenceExecutionConfig={"Mode": "Direct"},
ExecutionRoleArn=role,)
但我犯了这个错误。有线索吗？
---------------------------------------------------------------------------
ParamValidationError Traceback (most recent call last)
<ipython-input-96-c6e1810e5d03> in <module>
3 Containers=[pytorch_container, pytorch_container2],
4 InferenceExecutionConfig={"Mode": "Direct"},
----> 5 ExecutionRoleArn=role,
6 )
/opt/conda/lib/python3.7/site-packages/botocore/client.py in _api_call(self, *args, **kwargs)
355 use_fips_endpoint = client.meta.config.use_fips_endpoint
356 S3EndpointSetter(
--> 357 endpoint_resolver=self._endpoint_resolver,
358 region=client.meta.region_name,
359 s3_config=client.meta.config.s3,
/opt/conda/lib/python3.7/site-packages/botocore/client.py in _make_api_call(self, operation_name, api_params)
647 metadata=resolved,
648 signature_version=signature_version,
--> 649 )
650
651 def _resolve_endpoint_variant_config_var(self, config_var):
/opt/conda/lib/python3.7/site-packages/botocore/client.py in _convert_to_request_dict(self, api_params, operation_model, context)
695 ):
696 if endpoint_url is None:
--> 697 # Expand the default hostname URI template.
698 hostname = self.default_endpoint.format(
699 service=service_name, region=region_name
/opt/conda/lib/python3.7/site-packages/botocore/validate.py in serialize_to_request(self, parameters, operation_model)
295 # "type":"string",
296 # "min":1,
--> 297 # "max":256
298 # }
299 range_check(name, len(param), shape, 'invalid length', errors)
ParamValidationError: Parameter validation failed:
Unknown parameter in input: "InferenceExecutionConfig", must be one of: ModelName, PrimaryContainer, Containers, ExecutionRoleArn, Tags, VpcConfig, EnableNetworkIsolation提前感谢
如果您有类似的框架，从我看到的pytorch来看，我建议使用MME。MCE使用多个容器的示例可以在这里找到-https://github.com/aws-samples/sagemaker-hosting/blob/main/Advanced-Model-Deployment/Serial-Inference-Pipeline/Serial-Inference-Pipeline-with-Scikit-learn-and-Linear-Learner.ipynb