[sc-69429] SPSA keyword argument bug fix (#6027)

**Context:** `SPSAOptimizer`'s `step_and_cost` method was ignoring
keyword arguments in the objective function.

**Description of the Change:** `SPSAOptimizer.step_and_cost` no longer
ignores kwargs.

**Benefits:** Your keyword arguments are heard <3 

**Possible Drawbacks:** None

**Related github issue:**
#6028

doc/releases/changelog-dev.md

@@ -129,6 +129,9 @@
 
 <h3>Bug fixes 🐛</h3>
 
+* Fixed a bug in `qml.SPSAOptimizer` that ignored keyword arguments in the objective function.
+  [(#6027)](https://github.com/PennyLaneAI/pennylane/pull/6027)
+
 * `dynamic_one_shot` was broken for old-API devices since `override_shots` was deprecated.
   [(#6024)](https://github.com/PennyLaneAI/pennylane/pull/6024)
 

pennylane/optimize/spsa.py

@@ -194,6 +194,7 @@ def step_and_cost(self, objective_fn, *args, **kwargs):
             objective function output prior to the step.
         """
         g = self.compute_grad(objective_fn, args, kwargs)
+
         new_args = self.apply_grad(g, args)
 
         self.k += 1
@@ -270,7 +271,8 @@ def compute_grad(self, objective_fn, args, kwargs):
                 shots = Shots(objective_fn.device._raw_shot_sequence)  # pragma: no cover
             else:
                 shots = Shots(None)
-            if np.prod(objective_fn.func(*args).shape(objective_fn.device, shots)) > 1:
+
+            if np.prod(objective_fn.func(*args, **kwargs).shape(objective_fn.device, shots)) > 1:
                 raise ValueError(
                     "The objective function must be a scalar function for the gradient "
                     "to be computed."

tests/optimize/test_spsa.py

@@ -443,7 +443,7 @@ def cost(params):
 
     @pytest.mark.usefixtures("use_legacy_opmath")
     @pytest.mark.slow
-    def test_lighting_device_legacy_opmath(self):
+    def test_lightning_device_legacy_opmath(self):
         """Test SPSAOptimizer implementation with lightning.qubit device."""
         coeffs = [0.2, -0.543, 0.4514]
         obs = [
@@ -479,7 +479,7 @@ def cost_fun(params, num_qubits=1):
         assert energy < init_energy
 
     @pytest.mark.slow
-    def test_lighting_device(self):
+    def test_lightning_device(self):
         """Test SPSAOptimizer implementation with lightning.qubit device."""
         coeffs = [0.2, -0.543, 0.4514]
         obs = [
@@ -494,6 +494,9 @@ def test_lighting_device(self):
         @qml.qnode(dev)
         def cost_fun(params, num_qubits=1):
             qml.BasisState([1, 1, 0, 0], wires=range(num_qubits))
+
+            assert num_qubits == 4
+
             for i in range(num_qubits):
                 qml.Rot(*params[i], wires=0)
                 qml.CNOT(wires=[2, 3])