Sagemaker Inference for Tensorflow Base64 Input Error through API Gateway

Question

When I am trying to call my Sagemaker TF endpoint using API Gateway -> Lambda Func by passing a Base 64 String (an image) I am getting an unsupported string error. I also tried with application/Json but I am still getting the error. Need Suggestion. In Notebook Instance this is how my input looks:


input = {
  'instances': [{"b64": "iV"}]
}

In Lambda function I am doing this:


instance = [{"b64": "b64string"}]
pleasework=json.dumps({"instances": instance})
response = runtime.invoke_endpoint(EndpointName=ENDPOINT_NAME_BASE64,ContentType='string',Accept='string' ,Body=pleasework)

ERROR:
Inference Error:
An error occurred (ModelError) when calling the InvokeEndpoint operation: Received client error (415) from primary with message "{"error": "Unsupported Media Type: string"}".


Incase if I pass application/json I get this error:

Received client error (400) from primary with message "{    "error": "Failed to process element: 0 of 'instances' list. Error: INVALID_ARGUMENT: JSON Value: {
    "b64": "iV"
} Type: Object is not of expected type: uint8"}"

Answer

I would suggest testing invoking your model locally first and confirming what the input your model is expecting using the saved_model CLI. Kindly see this link: https://www.tensorflow.org/guide/saved_model#the_savedmodel_format_on_disk

Then when invoking the model confirm that instance is in the correct input format shape your model expects.

Then when invoking the model confirm that `instance` is in the correct input format shape your model expects.

Sagemaker Inference for Tensorflow Base64 Input Error through API Gateway

Relevant content