ONNX QCDQ Export Fails for Pytorch 2.9.0+cpu and dynamo=True

[JP-Amboage]

Trying to export a Quant model with export_onnx_qcdq and dynamo=True results in the following error when using Pytorch 2.9.0+cpu :

<class 'RuntimeError'>: Cannot access data pointer of Tensor (e.g. FakeTensor, FunctionalTensor). If you're using torch.compile/export/fx, it is likely that we are erroneously tracing into a custom kernel. To fix this, please wrap the custom kernel into an opaque custom op. Please see the following for details: https://pytorch.org/tutorials/advanced/custom_ops_landing_page.html

(Refer to the full stack trace above for more information.)

The same error is produced when dynamo is not explicitly set in the call to export_onnx_qcdq.

I provide the file needed to create the environment to reproduce the issue issue-env.yaml and a jupyter notebook issue.ipynb with the code that causes the issue and the error message visible in the output cells.

Finally here is a simple code snippet where the problem happens:

import torch
import brevitas.nn as qnn
from brevitas.export import export_onnx_qcdq

IN_CH = 3
OUT_CH = 128
BATCH_SIZE = 1

linear = qnn.QuantLinear(IN_CH, OUT_CH, bias=True)
inp = torch.randn(BATCH_SIZE, IN_CH)
path = 'quant_linear_qcdq.onnx'

exported_model = export_onnx_qcdq(linear, args=inp, export_path=path, opset_version=13)

[nickfraser]

Note that dynamo=True is the default for torch.onnx.export for PyTorch 2.9.