Hi all, I'm from the pytorch team and am looking into removing the experimental Named Tensor feature from pytorch entirely as a project to reduce overhead and code bloat. I'm looking into the implications other repos including this one.
From what I can tell, this repo references named tensors in the following way:
- Named tensor op overloads within the mirrored
native_functions.yaml
- A test file
test_namedtensor_xpu.py
- Some C++ usage of named tensor APIs
Q: Will this feature removal cause disruption I'm unaware of? Would you all accept a PR removing all of the above?
Hi all, I'm from the pytorch team and am looking into removing the experimental Named Tensor feature from pytorch entirely as a project to reduce overhead and code bloat. I'm looking into the implications other repos including this one.
From what I can tell, this repo references named tensors in the following way:
native_functions.yamltest_namedtensor_xpu.pyQ: Will this feature removal cause disruption I'm unaware of? Would you all accept a PR removing all of the above?