docs(autojac): Fix formatting (#575)

Author: ValerianRey

src/torchjd/autojac/_backward.py

@@ -26,9 +26,9 @@ def backward(
     ``.jac`` fields of the ``inputs``.
 
     :param tensors: The tensor or tensors to differentiate. Should be non-empty.
-    :param jac_tensors: The initial Jacobians to backpropagate, analog to the `grad_tensors`
-        parameter of `torch.autograd.backward`. If provided, it must have the same structure as
-        ``tensors`` and each tensor in ``jac_tensors`` must match the shape of the corresponding
+    :param jac_tensors: The initial Jacobians to backpropagate, analog to the ``grad_tensors``
+        parameter of :func:`torch.autograd.backward`. If provided, it must have the same structure
+        as ``tensors`` and each tensor in ``jac_tensors`` must match the shape of the corresponding
         tensor in ``tensors``, with an extra leading dimension representing the number of rows of
         the resulting Jacobian (e.g. the number of losses). All tensors in ``jac_tensors`` must
         have the same first dimension. If ``None``, defaults to the identity matrix. In this case,

src/torchjd/autojac/_jac.py

@@ -35,7 +35,7 @@ def jac(
         their ``requires_grad`` flag set to ``True``. If not provided, defaults to the leaf tensors
         that were used to compute the ``outputs`` parameter.
     :param jac_outputs: The initial Jacobians to backpropagate, analog to the ``grad_outputs``
-        parameter of ``torch.autograd.grad``. If provided, it must have the same structure as
+        parameter of :func:`torch.autograd.grad`. If provided, it must have the same structure as
         ``outputs`` and each tensor in ``jac_outputs`` must match the shape of the corresponding
         tensor in ``outputs``, with an extra leading dimension representing the number of rows of
         the resulting Jacobian (e.g. the number of losses). If ``None``, defaults to the identity

src/torchjd/autojac/_mtl_backward.py

@@ -50,7 +50,7 @@ def mtl_backward(
     :param features: The last shared representation used for all tasks, as given by the feature
         extractor. Should be non-empty.
     :param grad_tensors: The initial gradients to backpropagate, analog to the ``grad_tensors``
-        parameter of ``torch.autograd.backward``. If any of the ``tensors`` is non-scalar,
+        parameter of :func:`torch.autograd.backward`. If any of the ``tensors`` is non-scalar,
         ``grad_tensors`` must be provided, with the same length and shapes as ``tensors``.
         Otherwise, this parameter is not needed and will default to scalars of 1.
     :param tasks_params: The parameters of each task-specific head. Their ``requires_grad`` flags