test: Fix using wrong backward interface (#513)

Author: ValerianRey

tests/utils/forward_backwards.py

@@ -11,6 +11,7 @@
 from torchjd.aggregation import Aggregator, Weighting
 from torchjd.autogram import Engine
 from torchjd.autojac import backward
+from torchjd.autojac._jac_to_grad import jac_to_grad
 
 
 def autograd_forward_backward(
@@ -29,7 +30,8 @@ def autojac_forward_backward(
     aggregator: Aggregator,
 ) -> None:
     losses = forward_pass(model, inputs, loss_fn, reduce_to_vector)
-    backward(losses, aggregator=aggregator)
+    backward(losses)
+    jac_to_grad(model.parameters(), aggregator)
 
 
 def autograd_gramian_forward_backward(