[Bug]: OnnxModelHandlerNumpy broken with latest onnx version

[gariepyalex]

What happened?

There is a runtime error when loading a ONNX model with OnnxModelHandlerNumpy when using ONNX's latest version:

from apache_beam.ml.inference.onnx_inference import OnnxModelHandlerNumpy

model_handler = OnnxModelHandlerNumpy(
    "/tmp/model.onnx" # Also crashed with a remote model URI gs://...
)

Here is the error, which happens using the DirectRunner:

/3.10.14/lib/python3.10/site-packages/apache_beam/ml/inference/onnx_inference.py", line 119, in load_model
    model_proto_bytes = onnx._serialize(model_proto)
AttributeError: module 'onnx' has no attribute '_serialize' [while running 'RunInference/BeamML_RunInference']

pip freeze | grep onnx       
onnx==1.17.0
onnxruntime==1.19.2

pip freeze | grep beam
apache-beam==2.60.0

Issue Priority

Priority: 1 (data loss / total loss of function)

Issue Components

• Component: Python SDK
• Component: Java SDK
• Component: Go SDK
• Component: Typescript SDK
• Component: IO connector
• Component: Beam YAML
• Component: Beam examples
• Component: Beam playground
• Component: Beam katas
• Component: Website
• Component: Infrastructure
• Component: Spark Runner
• Component: Flink Runner
• Component: Samza Runner
• Component: Twister2 Runner
• Component: Hazelcast Jet Runner
• Component: Google Cloud Dataflow Runner

[gariepyalex]

The error goes away by downgrading to ONNX 1.14.1 (Aug 2023).

[damccorm]

Thanks for calling this out - working on a quick fix here - #33000

[gariepyalex]

Wow thanks @damccorm for the fast fix!