diff --git a/src/rbm/entity.py b/src/rbm/entity.py
index 82c9768..65a6c85 100644
--- a/src/rbm/entity.py
+++ b/src/rbm/entity.py
@@ -104,7 +104,7 @@ class Entity:
 		self.grad.b_v = momentum * self.grad.b_v + learning_rate * d_bv
 		self.grad.b_h = momentum * self.grad.b_h + learning_rate * d_bh
 		# compute L2 term and penalize cost function (d_whv)
-		l2_norm = np.sum(np.square(self.state.w_hv))/self.state.w_hv.shape[1]
+		l2_norm = 0.5*np.sum(np.square(self.state.w_hv))
 		l2_term = l2_lambda*l2_norm*self.state.w_hv
 		self.grad.w_hv = momentum * self.grad.w_hv + learning_rate * (d_whv-l2_term) - learning_rate * weight_decay * self.state.w_hv
 
diff --git a/src/tests/test_norbs.py b/src/tests/test_norbs.py
index a716a0f..afbafd2 100644
--- a/src/tests/test_norbs.py
+++ b/src/tests/test_norbs.py
@@ -16,7 +16,7 @@ class TestModel(Model):
 		else:
 			# Hidden binary
 			self.unit1 = Entity((96 * 96, 333), EntityParams(do_gaussian_visible=True, do_gaussian_hidden=False),
-								TrainingParams(learning_rate=0.001, momentum=0.9, num_epochs=1000, l2_lambda=0.4))
+								TrainingParams(learning_rate=0.001, momentum=0.9, num_epochs=1000, l2_lambda=0.01))
 
 	def forward(self, x: Mat):
 		x = self.unit1.forward(x)