pytorch/functorch

Issues

• Performance drop for the batching rule for aten::_sparse_mm

  #1075 opened 2 years ago by lrnzgiusti
  3

• Small difference between functorch grads and torch.autograd.grad

  #1144 opened a month ago by alanjeffares
  0

• item() support for vmap

  #1108 opened a year ago by baskuit
  2

• vmap does not support Tensor.clone()

  #1143 opened 2 months ago by JustinS6626
  0

• How to get only the last few layers' gradident?

  #1101 opened a year ago by pmzzs
  2

• Swapping 2 columns in a 2d tensor

  #1142 opened 3 months ago by Kreativshikkk
  0

• Per-sample-gradient: Get gradient 0 when using grad(params_tograd, params) with respect to part of model's parameters

  #1122 opened a year ago by Ancientshi
  1

• Will pmap be supported in functorh？

  #1099 opened a year ago by shixun404
  1

• wrapper->level().value() <= current_level INTERNAL ASSERT FAILED at "../aten/src/ATen/functorch/ADInterpreters.cpp":39

  #1131 opened a year ago by ElegantLin
  1

• Add vmap support for PyTorch operators

  #1088 opened a year ago by zou3519
  2

• vmap fails if your model includes full_backward_hook in pytorch2.0

  #1124 opened a year ago by AlphaBetaGamma96
  1

• Can I call torch.utils.data.WeightedRandomSampler inside vmap?

  #1123 opened a year ago by kwmaeng91
  1

• INTERNAL_ASSERT failed

  #1103 opened a year ago by Jeff09
  4

• fails to derive jacobian with functorch after operation torch.stft

  #1077 opened a year ago by lylyhan
  4

• Error about using a grad transform with in-place operation is inconsistent with and without DDP

  #1112 opened a year ago by XuchanBao
  1

• How to get the jacobian matrix in GCNs?

  #1113 opened a year ago by pcheng2
  0

• Use functional models inside usual nn.Module

  #1111 opened a year ago by subho406
  1

• Performance drop because of not yet implemented batching rule for bincount

  #1109 opened a year ago by IosiaLectus
  0

• Performance drop because of not yet implemented batching rule.

  #1069 opened a year ago by elientumba2019
  2

• vmap should be able to accept `None` in its out_dims argument

  #1082 opened a year ago by zou3519
  1

• vmap + GRU

  #1089 opened a year ago by belerico
  1

• Vmap and backward hook problem

  #1106 opened a year ago by pmzzs
  1

• RuntimeError: Batching rule not implemented for aten::is_same_size. We could not generate a fallback.

  #1104 opened a year ago by AlphaBetaGamma96
  0

• [Question] Packaging policy for `functorch` and `torch.func`

  #1102 opened a year ago by XuehaiPan
  5

• Unit Test Error When Testing vmap With Missing Module "autograd_function_db"

  #1098 opened a year ago by sunmorgan
  7

• [testing] Insufficient coverage in test suite

  #1090 opened a year ago by kshitij12345
  5

• UserWarning: There is a performance drop because we have not yet implemented the batching rule for `aten::greater_equal.Scalar`

  #1080 opened a year ago by carbocation
  3

• aten::all not implemented

  #1060 opened a year ago by iliTheFallen
  7

• Audit CompositeImplicitAutograd ops that do not have a batching rule, add them to BatchRulesDecomposition

  #1087 opened a year ago by zou3519
  1

• Make vmap tests use dtype `any_one`

  #1092 opened a year ago by samdow
  1

• Add pytorch 1.13.1 compatibility

  #1095 opened a year ago by scal444
  3

• batching over model parameters

  #1094 opened a year ago by LeanderK
  2

• `functorch.grad` slower than `torch.autograd.grad`

  #1085 opened a year ago by abdulfatir
  4

• [Question] Forward-mode AD `jvp` is slower than reverse-mode AD `vjp`

  #1083 opened a year ago by XuehaiPan
  2

• Entering and exiting a level should be done via the creation of Interpreter objects

  #1084 opened a year ago by zou3519
  0

• Installing functorch breaks torchaudio

  #1057 opened a year ago by dellis23
  11

• vmap should accept a dim_size=None argument

  #1081 opened a year ago by zou3519
  0

• make_fx fails with `jacfwd` (when used with torch.add(Tensor, Scalar))

  #1078 opened 2 years ago by kshitij12345
  1

• tree-based level system

  #1076 opened 2 years ago by zou3519
  2

• vmap equivalent for tensor[indices]

  #1074 opened 2 years ago by AlphaBetaGamma96
  1

• Different gradients for HyperNet training

  #1071 opened 2 years ago by bkoyuncu
  2

• Torch version constraints incompatible with NVIDIA NGC PyTorch image

  #1072 opened 2 years ago by timeatokai
  1

• Cannot build on Jetson platform

  #1065 opened 2 years ago by huyaoyu
  2

• Applying grad elementwise to tensors of arbitrary shape

  #1070 opened 2 years ago by EmilienDupont
  4

• dedup-error checking code (from structured kernels) with batching rules

  #1068 opened 2 years ago by zou3519
  4

• Unable to compute derivatives due to calling .item()

  #1066 opened 2 years ago by elientumba2019
  3

• installation issue

  #1062 opened 2 years ago by Tsingularity
  4

• runtime performance gain on model ensembling

  #1061 opened 2 years ago by xuyxu
  2

• Cuda Memory Overflow in Jacobian Computation

  #1058 opened 2 years ago by kayween
  6

• function(Jacobian)-dot-vector and vector-Jacobian-vector function

  #1056 opened 2 years ago by veya2ztn
  0