Yichen Yan

@XuehaiPan It's unexpected when we have `apache-tvm-ffi` in dependency but use a bundled lib. Space is not my main concern.

> Then you need to set an unreliable RPATH to `site-packages` I don't think that's necessary, when we have `import tvm_ffi` in python, `libtvm_ffi.so` would be loaded automatically.

> > > Then you need to set an unreliable RPATH to `site-packages` > > > > > > I don't think that's necessary, when we have `import tvm_ffi` in...

> `apache-tvm-ffi` claims it supports Python 3.8, but not tested. Checked a little bit, this ~not exactly a tvm-ffi issue. `ml_dtypes` is a optional dependencies:~ is a tvm-ffi issue that...

@justinchuby importing some submodule in torch might trigger importing `torch.onnx`, e.g.: ``` /mnt/c.py(53)() -> from torch._inductor.utils import maybe_profile /usr/local/lib/python3.10/site-packages/torch/_inductor/utils.py(49)() -> import torch._export /usr/local/lib/python3.10/site-packages/torch/_export/__init__.py(25)() -> import torch._dynamo /usr/local/lib/python3.10/site-packages/torch/_dynamo/__init__.py(2)() -> from ....

@pytorchbot merge

@pytorchbot rebase

@pytorchbot merge

unexpected imported `onnxscript` breaks some type annotation, I'm not sure about how the annotation part, trying to get more laziness to fix it.

> * [pull / linux-focal-py3.12-clang10 / test (default, 2, 4, linux.2xlarge)](https://github.com/pytorch/pytorch/actions/runs/10625230615/job/29455207085) > * [pull / linux-focal-py3.9-clang10-onnx / test (default, 2, 2, linux.2xlarge)](https://github.com/pytorch/pytorch/actions/runs/10625230615/job/29455236655) > * [pull / linux-jammy-py3.9-gcc11 / test (jit_legacy,...