TorchDynamo: always convert tensor to fake tensor at fake_mode path for ShapeProp (#90685)
This PR will fix https://github.com/pytorch/torchdynamo/issues/1978, for HF models, there is always report a ShapeProp error, the root cause is that we use fake tensor mode to do the ShapeProp, but for **torch.ones**, it always gets a none fake tensor and introduces an operation with non-fake tensors with fake tensors.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/90685
Approved by: https://github.com/ezyang, https://github.com/jansel