Use convert_attribute=True by default for onnx.save

[fxmarty]

As per title, see https://github.com/onnx/onnx/blob/b48d763a2d22574540f51b6bc5ce1a537fb65024/onnx/__init__.py#L310-L312

Have this as False may raise ValueError: The proto size is larger than the 2 GB limit. Please use save_as_external_data to save tensors separately from the model file. despite save_as_external_data=True.

Related: onnx/onnx#5949

We did not have errors previously as weights are Initializers, but Brevitas is using (not sure why) Constant for weights cc @Giuseppe5

[Giuseppe5]

I suspect the reason why we have constant is because the integer value of the weights is computed at runtime but the original weights are never "replaced" with their integer counterparts.

As far as ONNX export is concerned, those weights are constant values that are not part of the original model and they have no relations with the original weights.

In any case, thanks for fixing this!

[fxmarty]

I am not sure how torchscript -> onnx decides what is Initializer and what is Constant, to be honest.

[fxmarty]

Getting

  File "/home/felix/miniconda3/envs/fx/lib/python3.9/runpy.py", line 197, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/home/felix/miniconda3/envs/fx/lib/python3.9/runpy.py", line 87, in _run_code
    exec(code, run_globals)
  File "/home/felix/optimum/optimum/exporters/onnx/__main__.py", line 412, in <module>
    main()
  File "/home/felix/optimum/optimum/exporters/onnx/__main__.py", line 388, in main
    main_export(
  File "/home/felix/optimum/optimum/exporters/onnx/__main__.py", line 351, in main_export
    onnx_export_from_model(
  File "/home/felix/optimum/optimum/exporters/onnx/convert.py", line 1157, in onnx_export_from_model
    _, onnx_outputs = export_models(
  File "/home/felix/optimum/optimum/exporters/onnx/convert.py", line 768, in export_models
    export(
  File "/home/felix/optimum/optimum/exporters/onnx/convert.py", line 902, in export
    config.fix_dynamic_axes(output, device=device, input_shapes=input_shapes, dtype=dtype)
  File "/home/felix/optimum/optimum/exporters/onnx/base.py", line 307, in fix_dynamic_axes
    session = InferenceSession(model_path.as_posix(), providers=providers, sess_options=session_options)
  File "/home/felix/miniconda3/envs/fx/lib/python3.9/site-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 419, in __init__
    self._create_inference_session(providers, provider_options, disabled_optimizers)
  File "/home/felix/miniconda3/envs/fx/lib/python3.9/site-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 472, in _create_inference_session
    sess = C.InferenceSession(session_options, self._model_path, True, self._read_config_from_model)
onnxruntime.capi.onnxruntime_pybind11_state.Fail: [ONNXRuntimeError] : 1 : FAIL : Load model from /tmp/tmpo6_v2ag3/encoder_model.onnx failed:Node (/Unsqueeze_1) Op (Unsqueeze) [ShapeInferenceError] Cannot parse data from external tensors. Please load external data into raw data for tensor: /Constant_2_output_0

[fxmarty]

Related microsoft/onnxruntime#17088 microsoft/onnxruntime#11137

[fxmarty]

Merging as tests pass.