diff --git a/src/qiboml/models/encoding.py b/src/qiboml/models/encoding.py
index ff9f08d..b32aefa 100644
--- a/src/qiboml/models/encoding.py
+++ b/src/qiboml/models/encoding.py
@@ -62,7 +62,6 @@ def __call__(self, x: ndarray) -> Circuit:
                 f"Invalid input dimension {x.shape[-1]}, but the allocated qubits are {self.qubits}.",
             )
         circuit = self.circuit.copy()
-        ones = np.flatnonzero(x.ravel() == 1)
-        for bit in ones:
-            circuit.add(gates.X(self.qubits[bit]))
+        for qubit, bit in zip(self.qubits, x.ravel()):
+            circuit.add(gates.RX(qubit, theta=bit * np.pi, trainable=False))
         return circuit
diff --git a/tests/.#test_models_interfaces.py b/tests/.#test_models_interfaces.py
deleted file mode 120000
index 8d70c74..0000000
--- a/tests/.#test_models_interfaces.py
+++ /dev/null
@@ -1 +0,0 @@
-andrea@ubuntu-desktop.6447:1732264529
\ No newline at end of file
diff --git a/tests/test_models_encoding.py b/tests/test_models_encoding.py
index a2324de..d2b3dff 100644
--- a/tests/test_models_encoding.py
+++ b/tests/test_models_encoding.py
@@ -10,8 +10,8 @@ def test_binary_encoding_layer(backend):
     layer = ed.BinaryEncoding(nqubits, qubits=qubits)
     data = backend.cast(np.random.choice([0, 1], size=(len(qubits),)))
     c = layer(data)
-    indices = [gate.qubits[0] for gate in c.queue if gate.name == "x"]
-    assert [qubits[i] for i in np.flatnonzero(data == 1)] == indices
+    for bit, gate in zip(data, c.queue):
+        assert bit == gate.init_kwargs["theta"] / np.pi
     # test shape error
     with pytest.raises(RuntimeError):
         layer(backend.cast(np.random.choice([0, 1], size=(len(qubits) - 1,))))
diff --git a/tests/test_models_interfaces.py b/tests/test_models_interfaces.py
index b55629a..1618193 100644
--- a/tests/test_models_interfaces.py
+++ b/tests/test_models_interfaces.py
@@ -94,7 +94,7 @@ def random_tensor(frontend, shape, binary=False):
 
 
 def train_model(frontend, model, data, target):
-    max_epochs = 30
+    max_epochs = 10
     if frontend.__name__ == "qiboml.models.pytorch":
 
         optimizer = frontend.torch.optim.Adam(model.parameters())
@@ -168,7 +168,10 @@ def random_parameters(frontend, model):
     if frontend.__name__ == "qiboml.models.pytorch":
         new_params = {}
         for k, v in model.state_dict().items():
-            new_params.update({k: v + frontend.torch.randn(v.shape) / 2})
+            new_params.update(
+                {k: v + frontend.torch.randn(v.shape) / 5}
+            )  # perturbation of max +- 0.2
+            # of the original parameters
     elif frontend.__name__ == "qiboml.models.keras":
         new_params = [frontend.tf.random.uniform(model.get_weights()[0].shape)]
     return new_params
@@ -230,11 +233,11 @@ def test_encoding(backend, frontend, layer, seed):
     target = prepare_targets(frontend, q_model, data)
     backprop_test(frontend, q_model, data, target)
 
-    data = random_tensor(frontend, (100, 32))
+    data = random_tensor(frontend, (100, 4))
     model = build_sequential_model(
         frontend,
         [
-            build_linear_layer(frontend, 32, dim),
+            build_linear_layer(frontend, 4, dim),
             q_model,
             build_linear_layer(frontend, 2**nqubits, 1),
         ],
@@ -290,12 +293,12 @@ def test_decoding(backend, frontend, layer, seed, analytic):
     model = build_sequential_model(
         frontend,
         [
-            build_linear_layer(frontend, 32, dim),
+            build_linear_layer(frontend, 4, dim),
             q_model,
             build_linear_layer(frontend, q_model.output_shape[-1], 1),
         ],
     )
 
-    data = random_tensor(frontend, (100, 32))
+    data = random_tensor(frontend, (100, 4))
     target = prepare_targets(frontend, model, data)
     backprop_test(frontend, model, data, target)