Even though it is not documented, many users still use it. And it leads to many bugs in user code.
So we should remove it completely to prevent this.
The expected steps are:

• Add a new api to make a shallow copy that does not share version? Or a cat + unbind function that does not share version counter if that is enough.
• Remove the use of .data in all our internal code
  • benchmarks/
  • docs/source/scripts/build_activation_images.py and docs/source/notes/extending.rst
  • test/
  • torch/autograd/ Blocky by Remove use of .data in optimizers #30258 (comment) (not blocked ones were done in Remove some .data and clearly mark why we keep some #31479)
  • torch/cuda/ Use that requires the new api mentioned above.
  • torch/jit remove .data from torch/jit #31480
  • torch/nn Simple ones done in remove .data from torch/nn doc #31481 and remove simple .data from torch/nn #31482
  • torch/optim
  • torch/utils/tensorboard/
  • torch/tensor.py
  • ignored caffe2/
• Add warning to every call to .data
• Actually remove .data ?

cc @ezyang @ssnl @albanD @zou3519 @gqchen