albanD/subclass_zoo

Issues

• In complex numbers example, mm sample impl seems just regular complex number multiplication impl and not mm

  #53 opened 4 months ago by vadimkantorov
  1

• Question: apply trainable scale for qdq `linear` and `matmul`

  #51 opened 7 months ago by yiliu30
  3

• `torch.compile`'ed function inside `__torch_dispatch__`

  #47 opened a year ago by HanGuo97
  6

• `__torch_dispatch__` vs `__torch_function__`

  #46 opened a year ago by HanGuo97
  6

• What's the right way to use torch dispatch?

  #35 opened 2 years ago by lichao0613
  3

• "is a" tensor printing no longer works

  #30 opened 3 years ago by ezyang
  0

• How to retain the grad of via __torch_dispatch__ for torch.Tensor method

  #29 opened 3 years ago by wang-chen
  5

• Current "inheritance trivial tensor" doesn't seem to really work with in-place ops.

  #28 opened 3 years ago by Chillee
  2

• AttributeError: 'super' object has no attribute '__torch_dispatch__'

  #27 opened 3 years ago by wang-chen
  2

• Wrapper Tensor subclasses need to correctly propagate conjugate and neg bit

  #24 opened 3 years ago by zou3519
  0

• subclass.data = other_subclass is extremely dodgy

  #22 opened 3 years ago by ezyang
  1

• Compositional modes

  #11 opened 3 years ago by ezyang
  1

• __torch_function__ mode

  #12 opened 3 years ago by ezyang
  0

• Dispatcher-ify all operations on Tensor

  #21 opened 3 years ago by ezyang
  1

• Take the data in torchy typings and run it on the test suite

  #20 opened 3 years ago by ezyang
  0

• Automatically turn on TORCH_SHOW_CPP_STACKTRACES when using Python key

  #19 opened 3 years ago by ezyang
  0

• Simple functorch: cannot nest vmaps

  #16 opened 3 years ago by samdow
  0

• Example of Tensor Subclasses with nn.Modules?

  #17 opened 3 years ago by zou3519
  1

• Simple functorch: vmap should handle case where in_dims is None

  #15 opened 3 years ago by zou3519
  0

• Fix composite compliance

  #14 opened 3 years ago by zou3519
  0

• Per-backend conjugate keys

  #13 opened 3 years ago by ezyang
  0

• Schema access needs to get finished and then dogfooded to see if there are gaps

  #10 opened 3 years ago by ezyang
  1

• __torch_dispatch__ to __torch_function__ recompositions

  #9 opened 3 years ago by ezyang
  0

• Make it possible to construct tensor subclasses without having to always redefine __new__

  #8 opened 3 years ago by ezyang
  0

• Make dir work on torch.ops.aten

  #7 opened 3 years ago by ezyang
  0

• Once and for all figure out what the API status of disable torch dispatch/function is

  #6 opened 3 years ago by ezyang
  0

• Save and restore TLS has to also reset torch_function disabled state

  #5 opened 3 years ago by ezyang
  0

• Factory functions should be changed to CompositeExplicitAutograd

  #4 opened 3 years ago by ezyang
  0