Update on "Add log_sigmoid_backward forward-AD"

[ghstack-poisoned]
pytorch · Apr 17, 2023 · c41e998 · c41e998
1 parent 4f3def1
commit c41e998
Show file tree

Hide file tree

Showing 3 changed files with 4 additions and 0 deletions.
diff --git a/test/functorch/test_ops.py b/test/functorch/test_ops.py
@@ -1372,6 +1372,7 @@ def get_vjp(cotangents, *primals):
         xfail('renorm', ''),  # NYI: forward AD for renorm
         xfail('ormqr', ''),  # NYI: forward AD for ormqr
         xfail('nn.functional.multilabel_margin_loss', ''),  # NYI: multilabel_margin_loss_forward
+        xfail('nn.functional.soft_margin_loss', ''),  # NYI: forward-AD for soft_margin_loss_backward
         xfail('nn.functional.ctc_loss', ''),  # NYI: forward-AD for _ctc_loss
         xfail('nn.functional.pdist', ''),  # NYI: forward-AD with _pdist_forward
         skip('nn.functional.scaled_dot_product_attention', device_type='cuda'),

diff --git a/tools/autograd/derivatives.yaml b/tools/autograd/derivatives.yaml
@@ -1998,6 +1998,7 @@
 - name: log_sigmoid_forward(Tensor self) -> (Tensor output, Tensor buffer)
   self: log_sigmoid_backward(grad, self, buffer)
   output: log_sigmoid_backward(self_t.conj(), self_p, buffer).conj()
+  output_differentiability: [True, False]
 
 - name: _log_softmax(Tensor self, int dim, bool half_to_float) -> Tensor
   self: _log_softmax_backward_data(grad, result, dim, self.scalar_type())

diff --git a/torch/testing/_internal/common_methods_invocations.py b/torch/testing/_internal/common_methods_invocations.py
@@ -12526,6 +12526,7 @@ def reference_flatten(input, start_dim=0, end_dim=-1):
         dtypesIfCUDA=floating_types_and(torch.float16, torch.bfloat16),
         sample_inputs_func=sample_inputs_multilabel_soft_margin_loss,
         supports_forward_ad=True,
+        supports_fwgrad_bwgrad=True,
         decorators=(
             DecorateInfo(
                 toleranceOverride({torch.float32: tol(atol=1e-4, rtol=1e-4)}),
@@ -13132,6 +13133,7 @@ def reference_flatten(input, start_dim=0, end_dim=-1):
         supports_autograd=True,
         assert_autodiffed=False,
         supports_forward_ad=True,
+        supports_fwgrad_bwgrad=True,
         supports_gradgrad=True,
         # autodiff_nonfusible_nodes=["aten::log_sigmoid"],
         decorators=[