diff --git a/tests/models/test_modeling_common.py b/tests/models/test_modeling_common.py
index 9260f16caa..b0cf4a717d 100644
--- a/tests/models/test_modeling_common.py
+++ b/tests/models/test_modeling_common.py
@@ -221,7 +221,7 @@ class ModelTesterMixin:
             if isinstance(new_image, dict):
                 new_image = new_image.to_tuple()[0]
 
-        max_diff = (image - new_image).abs().sum().item()
+        max_diff = (image - new_image).abs().max().item()
         self.assertLessEqual(max_diff, 5e-5, "Models give different forward passes")
 
     def test_getattr_is_correct(self):
@@ -351,7 +351,7 @@ class ModelTesterMixin:
             if isinstance(new_image, dict):
                 new_image = new_image.to_tuple()[0]
 
-        max_diff = (image - new_image).abs().sum().item()
+        max_diff = (image - new_image).abs().max().item()
         self.assertLessEqual(max_diff, 5e-5, "Models give different forward passes")
 
     @require_torch_2
diff --git a/tests/models/test_models_unet_2d.py b/tests/models/test_models_unet_2d.py
index bb5335ca30..0d3e20cd6a 100644
--- a/tests/models/test_models_unet_2d.py
+++ b/tests/models/test_models_unet_2d.py
@@ -137,7 +137,7 @@ class UNetLDMModelTests(ModelTesterMixin, UNetTesterMixin, unittest.TestCase):
             model_accelerate.config.in_channels,
             model_accelerate.config.sample_size,
             model_accelerate.config.sample_size,
-            generator=torch.manual_seed(0),
+            generator=torch.Generator("cpu").manual_seed(0),
         )
         noise = noise.to(torch_device)
         time_step = torch.tensor([10] * noise.shape[0]).to(torch_device)
@@ -263,7 +263,7 @@ class NCSNppModelTests(ModelTesterMixin, UNetTesterMixin, unittest.TestCase):
 
         output_slice = output[0, -3:, -3:, -1].flatten().cpu()
         # fmt: off
-        expected_output_slice = torch.tensor([-4842.8691, -6499.6631, -3800.1953, -7978.2686, -10980.7129, -20028.8535, 8148.2822, 2342.2905, 567.7608])
+        expected_output_slice = torch.tensor([-4836.2178, -6487.1470, -3816.8196, -7964.9302, -10966.3037, -20043.5957, 8137.0513, 2340.3328, 544.6056])
         # fmt: on
 
         self.assertTrue(torch_all_close(output_slice, expected_output_slice, rtol=1e-2))
diff --git a/tests/models/test_models_unet_2d_condition.py b/tests/models/test_models_unet_2d_condition.py
index 84bd3ecb6b..6910a34196 100644
--- a/tests/models/test_models_unet_2d_condition.py
+++ b/tests/models/test_models_unet_2d_condition.py
@@ -726,8 +726,8 @@ class UNet2DConditionModelTests(ModelTesterMixin, UNetTesterMixin, unittest.Test
             model.disable_xformers_memory_efficient_attention()
             off_sample = model(**inputs_dict).sample
 
-        assert (sample - on_sample).abs().max() < 1e-4
-        assert (sample - off_sample).abs().max() < 1e-4
+        assert (sample - on_sample).abs().max() <= 5e-4
+        assert (sample - off_sample).abs().max() <= 5e-4
 
     def test_custom_diffusion_processors(self):
         # enable deterministic behavior for gradient checkpointing