[BUG] For model larger than 2GB: model with IR version >= 3 must specify opset_import for ONNX

[tp-nan]

Describe the bug
the model is llama2-7b (with MHA replaced by a simple local function like lambda q, k, v: return q). When the number of layers is set to 2, onnxsim works well. When number of layers is set to 32:

Simplifying...
Traceback (most recent call last):
  File "/usr/local/bin/onnxsim", line 8, in <module>
    sys.exit(main())
             ^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/onnxsim/onnx_simplifier.py", line 489, in main
    model_opt, check_ok = simplify(
                          ^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/onnxsim/onnx_simplifier.py", line 199, in simplify
    model_opt_bytes = C.simplify(
                      ^^^^^^^^^^^
RuntimeError: model with IR version >= 3 must specify opset_import for ONNX

onnxsim.version
0.4.36

torch.version
'2.7.0a0+7c8ec84dab.nv25.03'
onnxruntime.version
'1.18.0'