Enable onnx inlining in benchmark for >2GB models (#111867)
Summary:
X-link: https://github.com/pytorch/pytorch/pull/111867
Approved by: https://github.com/thiagocrepaldi
Reviewed By: izaitsevfb
Differential Revision: D50626483
fbshipit-source-id: dd45c7944829861e16271f3a02c121c33709ea96
