diff --git a/deel/torchlip/init.py b/deel/torchlip/init.py
index d4285b0..573cf29 100644
--- a/deel/torchlip/init.py
+++ b/deel/torchlip/init.py
@@ -24,8 +24,8 @@
 # rights reserved. DEEL is a research program operated by IVADO, IRT Saint Exupéry,
 # CRIAQ and ANITI - https://www.deel.ai/
 # =====================================================================================
-"""
-"""
+""" """
+import warnings
 import torch
 
 from .normalizers import bjorck_normalization
@@ -57,6 +57,9 @@ def spectral_(
         eps_spectral (float): stopping criterion of iterative power method
         maxiter_spectral (int): maximum number of iterations for the power iteration
     """
+    warnings.warn(
+        "spectral_ initialization is deprecated, use torch.nn.init.orthogonal_ instead"
+    )
     with torch.no_grad():
         tensor.copy_(
             spectral_normalization(
@@ -91,6 +94,9 @@ def bjorck_(
         maxiter_bjorck (int): maximum number of iterations for bjorck algorithm
         beta: Value to use for the :math:`\beta` parameter.
     """
+    warnings.warn(
+        "bjorck_ initialization is deprecated, use torch.nn.init.orthogonal_ instead"
+    )
     with torch.no_grad():
         spectral_tensor = spectral_normalization(
             tensor, None, eps=eps_spectral, maxiter=maxiter_spectral
diff --git a/tests/test_compute_layer_sv.py b/tests/test_compute_layer_sv.py
index 9feca85..1d830ae 100644
--- a/tests/test_compute_layer_sv.py
+++ b/tests/test_compute_layer_sv.py
@@ -24,8 +24,7 @@
 # rights reserved. DEEL is a research program operated by IVADO, IRT Saint Exupéry,
 # CRIAQ and ANITI - https://www.deel.ai/
 # =====================================================================================
-"""Tests for singular value computation (in compute_layer_sv.py)
-"""
+"""Tests for singular value computation (in compute_layer_sv.py)"""
 import os
 import pprint
 import pytest
diff --git a/tests/test_condense.py b/tests/test_condense.py
index be802e4..8777f92 100644
--- a/tests/test_condense.py
+++ b/tests/test_condense.py
@@ -301,9 +301,10 @@ def test_model(
     vanilla_loss2, vanilla_mse2 = uft.run_test(
         vanilla_model, test_dl, loss_fn, metrics, steps=10
     )
-    np.testing.assert_equal(
+    np.testing.assert_almost_equal(
         vanilla_mse,
         vanilla_mse2,
+        5,
         "exported model must be completely independent from original",
     )
 
diff --git a/tests/test_initializers.py b/tests/test_initializers.py
index 8a15c9b..e47888f 100644
--- a/tests/test_initializers.py
+++ b/tests/test_initializers.py
@@ -35,6 +35,10 @@
 )
 
 
+@pytest.mark.skipif(
+    hasattr(SpectralInitializer, "unavailable_class"),
+    reason="SpectralInitializer not available",
+)
 @pytest.mark.parametrize(
     "layer_type, layer_params,input_shape, orthogonal_test",
     [
diff --git a/tests/utils_framework.py b/tests/utils_framework.py
index a070176..4e326dd 100644
--- a/tests/utils_framework.py
+++ b/tests/utils_framework.py
@@ -64,7 +64,6 @@
     CategoricalHingeLoss,
 )
 
-from deel.torchlip.init import spectral_, bjorck_
 from deel.torchlip.normalizers import spectral_normalization
 from deel.torchlip.normalizers import bjorck_normalization
 from deel.torchlip.normalizers import DEFAULT_EPS_SPECTRAL
@@ -174,6 +173,7 @@ def __call__(self, **kwargs):
 Model = module_Unavailable_class
 compute_layer_sv = module_Unavailable_class
 OrthLinearRegularizer = module_Unavailable_class
+SpectralInitializer = module_Unavailable_class
 
 MODEL_PATH = "model.h5"
 LIP_LAYERS = "torchlip_layers"
@@ -591,13 +591,6 @@ def scaleDivAlpha(alpha):
     return 1.0 / (1 + 1.0 / alpha)
 
 
-def SpectralInitializer(eps_spectral, eps_bjorck):
-    if eps_bjorck is None:
-        return partial(spectral_, eps_spectral=eps_spectral)
-    else:
-        return partial(bjorck_, eps_spectral=eps_spectral, eps_bjorck=eps_bjorck)
-
-
 class tAdd(torch.nn.Module):
     def __init__(self):
         super(tAdd, self).__init__()