status: 500, error: InternalUncaught, message: UNKNOWN: Exception calling application: unsupported operand type(s) for *: 'TensorProto' and 'TensorProto'

[ravimadhusudhan8]

Created a Model -that was a simple calculator service - with standard add, subtract, multiply ... that take 2 parameters and output a single scalar.

The Model uploaded fine - no issues [the earlier suggestion helped].

Created an Application.

On testing the Application - got the error - status: 500, error: InternalUncaught, message: UNKNOWN: Exception calling application: unsupported operand type(s) for *: 'TensorProto' and 'TensorProto'

Attached screenshots -

[tidylobster]

Hi, @ravimadhusudhan8. ML Lambda communicates with the model via TensorProto objects. The message states that you cannot multiply two TensorProto objects together. You have to retrieve the content from the TensorProtos, perform add/multiply/substract operations on that content and then pack it back to the response TensorProto object. You can investigate error logs on your own by looking at the container logs.

[ravimadhusudhan8]

Got it!

Thanks

[ravimadhusudhan8]

Changed the code to this - based on func_main.py at hydro-serving-example/demos/stateful_lstm_example/pipeline/postprocessing/src

And using TensorProto objects as input and output as suggested. Attached are the docker logs and test app screenshots.

thanks

---

import math import hydro_serving_grpc as hs

def multiply(x, y): input_x = x.double_val input_y = y.double_val print(input_x, input_y) result = input_x * input_y tensor_result_output = hs.TensorProto( dtype=hs.DT_DOUBLE, double_val=result ) return tensor_result_output

---

Got this error message - status: 500, error: InternalUncaught, message: UNKNOWN: Exception calling application: unsupported operand type(s) for *: 'google.protobuf.pyext._message.RepeatedScalarContainer' and 'google.protobuf.pyext._message.RepeatedScalarContainer'

[tidylobster]

Still, you're operating on the object, that doesn't support * operand. This time it's a protobuf's RepeatedScalarContainer object, which is similar to list. In your case those would be a lists of length 1. So, the correct steps would be:

...
result = input_x[0] * input_y[0]
tensor_result_output = hs.TensorProto(
    dtype=hs.DT_DOUBLE,
    double_val=[result]
)
...

Note, that I've made a list from result and passed it to double_val key in the response tensor constructor.

For future reference, the documentation for the python bindings to protocol buffers explains this and other issues.

[ravimadhusudhan8]

Thanks for the link and suggestions.

Getting a different error with the changed code

import math import hydro_serving_grpc as hs def subtract(x, y): input_x = x.double_val input_y = y.double_val print(input_x[0], input_y[0]) result = input_x[0] - input_y[0] tensor_result_output = hs.TensorProto( dtype=hs.DT_DOUBLE, double_val=[result] ) return tensor_result_output

[tidylobster]

ML Lambda expects a hs.PredictResponse object to be returned from the python model. 4-th step here.

[ravimadhusudhan8]

Great that worked! You have been very responsive! Thanks!