VertexAI Model Prediction an Explanation returning empty error message

15 views Asked by At

Asking for a prediction with explaination from the VertexAI SDK, REST, and CLI return HTTP code 400 with no error message

From the REST API: https://-aiplatform.googleapis.com/v1/:explain the returned message is: {"error": "b'{"error": ""}'"}

The Python SDK for endpoint.explain(instances=instances, parameters=parameters) method returns this error: InvalidArgument: 400 {"error": "b'{"error": ""}'"}

gcloud ai endpoints explain --region= --json-request=/request.json ERROR: (gcloud.beta.ai.endpoints.explain) HTTPError 400: {"error": "b'{"error": ""}'"}

To ensure that I hadn't made a mistake, I tried the predict method. The SDK method aiplatform.Endpoint.predict(instances=instances, parameters=parameters) returns a valid response

It appears that the REST method is correct based upon documentation here.

It also appears that only a correct instances argument is needed for the explain method based upon documentation here, but I did try it without parameters, and with including a deployed_model_id and timeout.

I think the CLI is correct, though I am not as versed with using it as REST or the SDK based upon the documentation here

I would have expected these methods to work, or at the very least provide an information about why the request was incorrect.

0

There are 0 answers