Fix typo and format BasicTransformerBlock attributes (#2953)

* ⚙️chore(train_controlnet) fix typo in logger message * ⚙️chore(models) refactor modules order; make them the same as calling order When printing the BasicTransformerBlock to stdout, I think it's crucial that the attributes order are shown in proper order. And also previously the "3. Feed Forward" comment was not making sense. It should have been close to self.ff but it's instead next to self.norm3 * correct many tests * remove bogus file * make style * correct more tests * finish tests * fix one more * make style * make unclip deterministic * ⚙️chore(models/attention) reorganize comments in BasicTransformerBlock class --------- Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
2026-01-27 17:22:53 +03:00 · 2023-04-12 05:31:05 +07:00
parent c6180a311c
commit 52c4d32d41
25 changed files with 87 additions and 154 deletions
--- a/examples/controlnet/train_controlnet.py
+++ b/examples/controlnet/train_controlnet.py
@@ -577,7 +577,7 @@ def make_train_dataset(args, tokenizer, accelerator):

    if args.conditioning_image_column is None:
        conditioning_image_column = column_names[2]
-        logger.info(f"conditioning image column defaulting to {caption_column}")
+        logger.info(f"conditioning image column defaulting to {conditioning_image_column}")
    else:
        conditioning_image_column = args.conditioning_image_column
        if conditioning_image_column not in column_names:
--- a/src/diffusers/models/attention.py
+++ b/src/diffusers/models/attention.py
@@ -224,7 +224,14 @@ class BasicTransformerBlock(nn.Module):
                f" define `num_embeds_ada_norm` if setting `norm_type` to {norm_type}."
            )

+        # Define 3 blocks. Each block has its own normalization layer.
        # 1. Self-Attn
+        if self.use_ada_layer_norm:
+            self.norm1 = AdaLayerNorm(dim, num_embeds_ada_norm)
+        elif self.use_ada_layer_norm_zero:
+            self.norm1 = AdaLayerNormZero(dim, num_embeds_ada_norm)
+        else:
+            self.norm1 = nn.LayerNorm(dim, elementwise_affine=norm_elementwise_affine)
        self.attn1 = Attention(
            query_dim=dim,
            heads=num_attention_heads,
@@ -235,10 +242,16 @@ class BasicTransformerBlock(nn.Module):
            upcast_attention=upcast_attention,
        )

-        self.ff = FeedForward(dim, dropout=dropout, activation_fn=activation_fn, final_dropout=final_dropout)
-
        # 2. Cross-Attn
        if cross_attention_dim is not None or double_self_attention:
+            # We currently only use AdaLayerNormZero for self attention where there will only be one attention block.
+            # I.e. the number of returned modulation chunks from AdaLayerZero would not make sense if returned during
+            # the second cross attention block.
+            self.norm2 = (
+                AdaLayerNorm(dim, num_embeds_ada_norm)
+                if self.use_ada_layer_norm
+                else nn.LayerNorm(dim, elementwise_affine=norm_elementwise_affine)
+            )
            self.attn2 = Attention(
                query_dim=dim,
                cross_attention_dim=cross_attention_dim if not double_self_attention else None,
@@ -248,30 +261,13 @@ class BasicTransformerBlock(nn.Module):
                bias=attention_bias,
                upcast_attention=upcast_attention,
            )  # is self-attn if encoder_hidden_states is none
-        else:
-            self.attn2 = None
-
-        if self.use_ada_layer_norm:
-            self.norm1 = AdaLayerNorm(dim, num_embeds_ada_norm)
-        elif self.use_ada_layer_norm_zero:
-            self.norm1 = AdaLayerNormZero(dim, num_embeds_ada_norm)
-        else:
-            self.norm1 = nn.LayerNorm(dim, elementwise_affine=norm_elementwise_affine)
-
-        if cross_attention_dim is not None or double_self_attention:
-            # We currently only use AdaLayerNormZero for self attention where there will only be one attention block.
-            # I.e. the number of returned modulation chunks from AdaLayerZero would not make sense if returned during
-            # the second cross attention block.
-            self.norm2 = (
-                AdaLayerNorm(dim, num_embeds_ada_norm)
-                if self.use_ada_layer_norm
-                else nn.LayerNorm(dim, elementwise_affine=norm_elementwise_affine)
-            )
        else:
            self.norm2 = None
+            self.attn2 = None

        # 3. Feed-forward
        self.norm3 = nn.LayerNorm(dim, elementwise_affine=norm_elementwise_affine)
+        self.ff = FeedForward(dim, dropout=dropout, activation_fn=activation_fn, final_dropout=final_dropout)

    def forward(
        self,
@@ -283,6 +279,8 @@ class BasicTransformerBlock(nn.Module):
        cross_attention_kwargs=None,
        class_labels=None,
    ):
+        # Notice that normalization is always applied before the real computation in the following blocks.
+        # 1. Self-Attention
        if self.use_ada_layer_norm:
            norm_hidden_states = self.norm1(hidden_states, timestep)
        elif self.use_ada_layer_norm_zero:
@@ -292,7 +290,6 @@ class BasicTransformerBlock(nn.Module):
        else:
            norm_hidden_states = self.norm1(hidden_states)

-        # 1. Self-Attention
        cross_attention_kwargs = cross_attention_kwargs if cross_attention_kwargs is not None else {}
        attn_output = self.attn1(
            norm_hidden_states,
@@ -304,6 +301,7 @@ class BasicTransformerBlock(nn.Module):
            attn_output = gate_msa.unsqueeze(1) * attn_output
        hidden_states = attn_output + hidden_states

+        # 2. Cross-Attention
        if self.attn2 is not None:
            norm_hidden_states = (
                self.norm2(hidden_states, timestep) if self.use_ada_layer_norm else self.norm2(hidden_states)
@@ -311,7 +309,6 @@ class BasicTransformerBlock(nn.Module):
            # TODO (Birch-San): Here we should prepare the encoder_attention mask correctly
            # prepare attention mask here

-            # 2. Cross-Attention
            attn_output = self.attn2(
                norm_hidden_states,
                encoder_hidden_states=encoder_hidden_states,
--- a/tests/pipelines/altdiffusion/test_alt_diffusion_img2img.py
+++ b/tests/pipelines/altdiffusion/test_alt_diffusion_img2img.py
@@ -166,7 +166,7 @@ class AltDiffusionImg2ImgPipelineFastTests(unittest.TestCase):
        image_from_tuple_slice = image_from_tuple[0, -3:, -3:, -1]

        assert image.shape == (1, 32, 32, 3)
-        expected_slice = np.array([0.4115, 0.3870, 0.4089, 0.4807, 0.4668, 0.4144, 0.4151, 0.4721, 0.4569])
+        expected_slice = np.array([0.4427, 0.3731, 0.4249, 0.4941, 0.4546, 0.4148, 0.4193, 0.4666, 0.4499])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 5e-3
        assert np.abs(image_from_tuple_slice.flatten() - expected_slice).max() < 5e-3
--- a/tests/pipelines/dit/test_dit.py
+++ b/tests/pipelines/dit/test_dit.py
@@ -92,7 +92,7 @@ class DiTPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
        image_slice = image[0, -3:, -3:, -1]

        self.assertEqual(image.shape, (1, 16, 16, 3))
-        expected_slice = np.array([0.4380, 0.4141, 0.5159, 0.0000, 0.4282, 0.6680, 0.5485, 0.2545, 0.6719])
+        expected_slice = np.array([0.2946, 0.6601, 0.4329, 0.3296, 0.4144, 0.5319, 0.7273, 0.5013, 0.4457])
        max_diff = np.abs(image_slice.flatten() - expected_slice).max()
        self.assertLessEqual(max_diff, 1e-3)

--- a/tests/pipelines/latent_diffusion/test_latent_diffusion.py
+++ b/tests/pipelines/latent_diffusion/test_latent_diffusion.py
@@ -125,7 +125,7 @@ class LDMTextToImagePipelineFastTests(PipelineTesterMixin, unittest.TestCase):
        image_slice = image[0, -3:, -3:, -1]

        assert image.shape == (1, 16, 16, 3)
-        expected_slice = np.array([0.59450, 0.64078, 0.55509, 0.51229, 0.69640, 0.36960, 0.59296, 0.60801, 0.49332])
+        expected_slice = np.array([0.6101, 0.6156, 0.5622, 0.4895, 0.6661, 0.3804, 0.5748, 0.6136, 0.5014])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-3

--- a/tests/pipelines/paint_by_example/test_paint_by_example.py
+++ b/tests/pipelines/paint_by_example/test_paint_by_example.py
@@ -129,7 +129,7 @@ class PaintByExamplePipelineFastTests(PipelineTesterMixin, unittest.TestCase):
        image_slice = image[0, -3:, -3:, -1]

        assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.4701, 0.5555, 0.3994, 0.5107, 0.5691, 0.4517, 0.5125, 0.4769, 0.4539])
+        expected_slice = np.array([0.4686, 0.5687, 0.4007, 0.5218, 0.5741, 0.4482, 0.4940, 0.4629, 0.4503])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

--- a/tests/pipelines/semantic_stable_diffusion/test_semantic_diffusion.py
+++ b/tests/pipelines/semantic_stable_diffusion/test_semantic_diffusion.py
@@ -154,7 +154,7 @@ class SafeDiffusionPipelineFastTests(unittest.TestCase):
        image_from_tuple_slice = image_from_tuple[0, -3:, -3:, -1]

        assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.5644, 0.6018, 0.4799, 0.5267, 0.5585, 0.4641, 0.516, 0.4964, 0.4792])
+        expected_slice = np.array([0.5753, 0.6114, 0.5001, 0.5034, 0.5470, 0.4729, 0.4971, 0.4867, 0.4867])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
        assert np.abs(image_from_tuple_slice.flatten() - expected_slice).max() < 1e-2
@@ -200,7 +200,7 @@ class SafeDiffusionPipelineFastTests(unittest.TestCase):
        image_from_tuple_slice = image_from_tuple[0, -3:, -3:, -1]

        assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.5095, 0.5674, 0.4668, 0.5126, 0.5697, 0.4675, 0.5278, 0.4964, 0.4945])
+        expected_slice = np.array([0.5122, 0.5712, 0.4825, 0.5053, 0.5646, 0.4769, 0.5179, 0.4894, 0.4994])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
        assert np.abs(image_from_tuple_slice.flatten() - expected_slice).max() < 1e-2
--- a/tests/pipelines/stable_diffusion/test_stable_diffusion.py
+++ b/tests/pipelines/stable_diffusion/test_stable_diffusion.py
@@ -135,7 +135,7 @@ class StableDiffusionPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
        image_slice = image[0, -3:, -3:, -1]

        assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.5643, 0.6017, 0.4799, 0.5267, 0.5584, 0.4641, 0.5159, 0.4963, 0.4791])
+        expected_slice = np.array([0.5756, 0.6118, 0.5005, 0.5041, 0.5471, 0.4726, 0.4976, 0.4865, 0.4864])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

@@ -282,7 +282,7 @@ class StableDiffusionPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
        image_slice = image[0, -3:, -3:, -1]

        assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.5094, 0.5674, 0.4667, 0.5125, 0.5696, 0.4674, 0.5277, 0.4964, 0.4945])
+        expected_slice = np.array([0.5122, 0.5712, 0.4825, 0.5053, 0.5646, 0.4769, 0.5179, 0.4894, 0.4994])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

@@ -322,19 +322,7 @@ class StableDiffusionPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
        image_slice = image[0, -3:, -3:, -1]

        assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array(
-            [
-                0.47082293033599854,
-                0.5371589064598083,
-                0.4562119245529175,
-                0.5220914483070374,
-                0.5733777284622192,
-                0.4795039892196655,
-                0.5465868711471558,
-                0.5074326395988464,
-                0.5042197108268738,
-            ]
-        )
+        expected_slice = np.array([0.4873, 0.5443, 0.4845, 0.5004, 0.5549, 0.4850, 0.5191, 0.4941, 0.5065])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

@@ -353,19 +341,7 @@ class StableDiffusionPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
        image_slice = image[0, -3:, -3:, -1]

        assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array(
-            [
-                0.4707113206386566,
-                0.5372191071510315,
-                0.4563021957874298,
-                0.5220003724098206,
-                0.5734264850616455,
-                0.4794946610927582,
-                0.5463782548904419,
-                0.5074145197868347,
-                0.504422664642334,
-            ]
-        )
+        expected_slice = np.array([0.4872, 0.5444, 0.4846, 0.5003, 0.5549, 0.4850, 0.5189, 0.4941, 0.5067])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

@@ -384,19 +360,7 @@ class StableDiffusionPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
        image_slice = image[0, -3:, -3:, -1]

        assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array(
-            [
-                0.47082313895225525,
-                0.5371587872505188,
-                0.4562119245529175,
-                0.5220913887023926,
-                0.5733776688575745,
-                0.47950395941734314,
-                0.546586811542511,
-                0.5074326992034912,
-                0.5042197108268738,
-            ]
-        )
+        expected_slice = np.array([0.4873, 0.5443, 0.4845, 0.5004, 0.5549, 0.4850, 0.5191, 0.4941, 0.5065])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

@@ -468,19 +432,7 @@ class StableDiffusionPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
        image_slice = image[0, -3:, -3:, -1]

        assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array(
-            [
-                0.5108221173286438,
-                0.5688379406929016,
-                0.4685141146183014,
-                0.5098261833190918,
-                0.5657756328582764,
-                0.4631010890007019,
-                0.5226285457611084,
-                0.49129390716552734,
-                0.4899061322212219,
-            ]
-        )
+        expected_slice = np.array([0.5114, 0.5706, 0.4772, 0.5028, 0.5637, 0.4732, 0.5169, 0.4881, 0.4977])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

--- a/tests/pipelines/stable_diffusion/test_stable_diffusion_image_variation.py
+++ b/tests/pipelines/stable_diffusion/test_stable_diffusion_image_variation.py
@@ -119,7 +119,7 @@ class StableDiffusionImageVariationPipelineFastTests(PipelineTesterMixin, unitte
        image_slice = image[0, -3:, -3:, -1]

        assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.5167, 0.5746, 0.4835, 0.4914, 0.5605, 0.4691, 0.5201, 0.4898, 0.4958])
+        expected_slice = np.array([0.5239, 0.5723, 0.4796, 0.5049, 0.5550, 0.4685, 0.5329, 0.4891, 0.4921])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-3

@@ -139,7 +139,7 @@ class StableDiffusionImageVariationPipelineFastTests(PipelineTesterMixin, unitte
        image_slice = image[-1, -3:, -3:, -1]

        assert image.shape == (2, 64, 64, 3)
-        expected_slice = np.array([0.6568, 0.5470, 0.5684, 0.5444, 0.5945, 0.6221, 0.5508, 0.5531, 0.5263])
+        expected_slice = np.array([0.6892, 0.5637, 0.5836, 0.5771, 0.6254, 0.6409, 0.5580, 0.5569, 0.5289])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-3

--- a/tests/pipelines/stable_diffusion/test_stable_diffusion_img2img.py
+++ b/tests/pipelines/stable_diffusion/test_stable_diffusion_img2img.py
@@ -138,7 +138,7 @@ class StableDiffusionImg2ImgPipelineFastTests(PipelineTesterMixin, unittest.Test
        image_slice = image[0, -3:, -3:, -1]

        assert image.shape == (1, 32, 32, 3)
-        expected_slice = np.array([0.4492, 0.3865, 0.4222, 0.5854, 0.5139, 0.4379, 0.4193, 0.48, 0.4218])
+        expected_slice = np.array([0.4555, 0.3216, 0.4049, 0.4620, 0.4618, 0.4126, 0.4122, 0.4629, 0.4579])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-3

@@ -157,7 +157,7 @@ class StableDiffusionImg2ImgPipelineFastTests(PipelineTesterMixin, unittest.Test
        image_slice = image[0, -3:, -3:, -1]

        assert image.shape == (1, 32, 32, 3)
-        expected_slice = np.array([0.4065, 0.3783, 0.4050, 0.5266, 0.4781, 0.4252, 0.4203, 0.4692, 0.4365])
+        expected_slice = np.array([0.4593, 0.3408, 0.4232, 0.4749, 0.4476, 0.4115, 0.4357, 0.4733, 0.4663])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-3

@@ -176,7 +176,7 @@ class StableDiffusionImg2ImgPipelineFastTests(PipelineTesterMixin, unittest.Test
        image_slice = image[-1, -3:, -3:, -1]

        assert image.shape == (2, 32, 32, 3)
-        expected_slice = np.array([0.5144, 0.4447, 0.4735, 0.6676, 0.5526, 0.5454, 0.645, 0.5149, 0.4689])
+        expected_slice = np.array([0.4241, 0.5576, 0.5711, 0.4792, 0.4311, 0.5952, 0.5827, 0.5138, 0.5109])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-3

@@ -196,7 +196,7 @@ class StableDiffusionImg2ImgPipelineFastTests(PipelineTesterMixin, unittest.Test
        image_slice = image[0, -3:, -3:, -1]

        assert image.shape == (1, 32, 32, 3)
-        expected_slice = np.array([0.4367, 0.4986, 0.4372, 0.6706, 0.5665, 0.444, 0.5864, 0.6019, 0.5203])
+        expected_slice = np.array([0.4398, 0.4949, 0.4337, 0.6580, 0.5555, 0.4338, 0.5769, 0.5955, 0.5175])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-3

--- a/tests/pipelines/stable_diffusion/test_stable_diffusion_instruction_pix2pix.py
+++ b/tests/pipelines/stable_diffusion/test_stable_diffusion_instruction_pix2pix.py
@@ -124,7 +124,7 @@ class StableDiffusionInstructPix2PixPipelineFastTests(PipelineTesterMixin, unitt
        image = sd_pipe(**inputs).images
        image_slice = image[0, -3:, -3:, -1]
        assert image.shape == (1, 32, 32, 3)
-        expected_slice = np.array([0.7318, 0.3723, 0.4662, 0.623, 0.5770, 0.5014, 0.4281, 0.5550, 0.4813])
+        expected_slice = np.array([0.7526, 0.3750, 0.4547, 0.6117, 0.5866, 0.5016, 0.4327, 0.5642, 0.4815])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-3

@@ -142,7 +142,7 @@ class StableDiffusionInstructPix2PixPipelineFastTests(PipelineTesterMixin, unitt
        image_slice = image[0, -3:, -3:, -1]

        assert image.shape == (1, 32, 32, 3)
-        expected_slice = np.array([0.7323, 0.3688, 0.4611, 0.6255, 0.5746, 0.5017, 0.433, 0.5553, 0.4827])
+        expected_slice = np.array([0.7511, 0.3642, 0.4553, 0.6236, 0.5797, 0.5013, 0.4343, 0.5611, 0.4831])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-3

@@ -165,7 +165,7 @@ class StableDiffusionInstructPix2PixPipelineFastTests(PipelineTesterMixin, unitt
        image_slice = image[-1, -3:, -3:, -1]

        assert image.shape == (2, 32, 32, 3)
-        expected_slice = np.array([0.606, 0.5712, 0.5099, 0.598, 0.5805, 0.7205, 0.6793, 0.554, 0.5607])
+        expected_slice = np.array([0.5812, 0.5748, 0.5222, 0.5908, 0.5695, 0.7174, 0.6804, 0.5523, 0.5579])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-3

@@ -187,7 +187,7 @@ class StableDiffusionInstructPix2PixPipelineFastTests(PipelineTesterMixin, unitt
        print(",".join([str(x) for x in slice]))

        assert image.shape == (1, 32, 32, 3)
-        expected_slice = np.array([0.726, 0.3902, 0.4868, 0.585, 0.5672, 0.511, 0.3906, 0.551, 0.4846])
+        expected_slice = np.array([0.7417, 0.3842, 0.4732, 0.5776, 0.5891, 0.5139, 0.4052, 0.5673, 0.4986])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-3

--- a/tests/pipelines/stable_diffusion/test_stable_diffusion_model_editing.py
+++ b/tests/pipelines/stable_diffusion/test_stable_diffusion_model_editing.py
@@ -118,9 +118,7 @@ class StableDiffusionModelEditingPipelineFastTests(PipelineTesterMixin, unittest
        image_slice = image[0, -3:, -3:, -1]
        assert image.shape == (1, 64, 64, 3)

-        expected_slice = np.array(
-            [0.5217179, 0.50658035, 0.5003239, 0.41109088, 0.3595158, 0.46607107, 0.5323504, 0.5335255, 0.49187922]
-        )
+        expected_slice = np.array([0.4755, 0.5132, 0.4976, 0.3904, 0.3554, 0.4765, 0.5139, 0.5158, 0.4889])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

@@ -139,9 +137,7 @@ class StableDiffusionModelEditingPipelineFastTests(PipelineTesterMixin, unittest

        assert image.shape == (1, 64, 64, 3)

-        expected_slice = np.array(
-            [0.546259, 0.5108156, 0.50897664, 0.41931948, 0.3748669, 0.4669299, 0.5427151, 0.54561913, 0.49353]
-        )
+        expected_slice = np.array([0.4992, 0.5101, 0.5004, 0.3949, 0.3604, 0.4735, 0.5216, 0.5204, 0.4913])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

@@ -161,9 +157,7 @@ class StableDiffusionModelEditingPipelineFastTests(PipelineTesterMixin, unittest

        assert image.shape == (1, 64, 64, 3)

-        expected_slice = np.array(
-            [0.47106352, 0.53579676, 0.45798016, 0.514294, 0.56856745, 0.4788605, 0.54380214, 0.5046455, 0.50404465]
-        )
+        expected_slice = np.array([0.4747, 0.5372, 0.4779, 0.4982, 0.5543, 0.4816, 0.5238, 0.4904, 0.5027])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

--- a/tests/pipelines/stable_diffusion/test_stable_diffusion_panorama.py
+++ b/tests/pipelines/stable_diffusion/test_stable_diffusion_panorama.py
@@ -119,7 +119,7 @@ class StableDiffusionPanoramaPipelineFastTests(PipelineTesterMixin, unittest.Tes
        image_slice = image[0, -3:, -3:, -1]
        assert image.shape == (1, 64, 64, 3)

-        expected_slice = np.array([0.5101, 0.5006, 0.4962, 0.3995, 0.3501, 0.4632, 0.5339, 0.525, 0.4878])
+        expected_slice = np.array([0.4794, 0.5084, 0.4992, 0.3941, 0.3555, 0.4754, 0.5248, 0.5224, 0.4839])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

@@ -138,7 +138,7 @@ class StableDiffusionPanoramaPipelineFastTests(PipelineTesterMixin, unittest.Tes

        assert image.shape == (1, 64, 64, 3)

-        expected_slice = np.array([0.5326, 0.5009, 0.5074, 0.4133, 0.371, 0.464, 0.5432, 0.5429, 0.4896])
+        expected_slice = np.array([0.5029, 0.5075, 0.5002, 0.3965, 0.3584, 0.4746, 0.5271, 0.5273, 0.4877])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

@@ -158,9 +158,7 @@ class StableDiffusionPanoramaPipelineFastTests(PipelineTesterMixin, unittest.Tes

        assert image.shape == (1, 64, 64, 3)

-        expected_slice = np.array(
-            [0.48235387, 0.5423796, 0.46016198, 0.5377287, 0.5803722, 0.4876525, 0.5515428, 0.5045897, 0.50709957]
-        )
+        expected_slice = np.array([0.4934, 0.5455, 0.4847, 0.5022, 0.5572, 0.4833, 0.5207, 0.4952, 0.5051])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

--- a/tests/pipelines/stable_diffusion/test_stable_diffusion_pix2pix_zero.py
+++ b/tests/pipelines/stable_diffusion/test_stable_diffusion_pix2pix_zero.py
@@ -133,7 +133,7 @@ class StableDiffusionPix2PixZeroPipelineFastTests(PipelineTesterMixin, unittest.
        image = sd_pipe(**inputs).images
        image_slice = image[0, -3:, -3:, -1]
        assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.5184, 0.503, 0.4917, 0.4022, 0.3455, 0.464, 0.5324, 0.5323, 0.4894])
+        expected_slice = np.array([0.4863, 0.5053, 0.5033, 0.4007, 0.3571, 0.4768, 0.5176, 0.5277, 0.4940])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-3

@@ -151,7 +151,7 @@ class StableDiffusionPix2PixZeroPipelineFastTests(PipelineTesterMixin, unittest.
        image_slice = image[0, -3:, -3:, -1]

        assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.5464, 0.5072, 0.5012, 0.4124, 0.3624, 0.466, 0.5413, 0.5468, 0.4927])
+        expected_slice = np.array([0.5177, 0.5097, 0.5047, 0.4076, 0.3667, 0.4767, 0.5238, 0.5307, 0.4958])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-3

@@ -170,7 +170,7 @@ class StableDiffusionPix2PixZeroPipelineFastTests(PipelineTesterMixin, unittest.
        image_slice = image[0, -3:, -3:, -1]

        assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.5114, 0.5051, 0.5222, 0.5279, 0.5037, 0.5156, 0.4604, 0.4966, 0.504])
+        expected_slice = np.array([0.5421, 0.5525, 0.6085, 0.5279, 0.4658, 0.5317, 0.4418, 0.4815, 0.5132])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-3

@@ -187,7 +187,7 @@ class StableDiffusionPix2PixZeroPipelineFastTests(PipelineTesterMixin, unittest.
        image_slice = image[0, -3:, -3:, -1]

        assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.5185, 0.5027, 0.492, 0.401, 0.3445, 0.464, 0.5321, 0.5327, 0.4892])
+        expected_slice = np.array([0.4861, 0.5053, 0.5038, 0.3994, 0.3562, 0.4768, 0.5172, 0.5280, 0.4938])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-3

--- a/tests/pipelines/stable_diffusion_2/test_stable_diffusion.py
+++ b/tests/pipelines/stable_diffusion_2/test_stable_diffusion.py
@@ -134,7 +134,7 @@ class StableDiffusion2PipelineFastTests(PipelineTesterMixin, unittest.TestCase):
        image_slice = image[0, -3:, -3:, -1]

        assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.5649, 0.6022, 0.4804, 0.5270, 0.5585, 0.4643, 0.5159, 0.4963, 0.4793])
+        expected_slice = np.array([0.5753, 0.6113, 0.5005, 0.5036, 0.5464, 0.4725, 0.4982, 0.4865, 0.4861])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

@@ -151,7 +151,7 @@ class StableDiffusion2PipelineFastTests(PipelineTesterMixin, unittest.TestCase):
        image_slice = image[0, -3:, -3:, -1]

        assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.5099, 0.5677, 0.4671, 0.5128, 0.5697, 0.4676, 0.5277, 0.4964, 0.4946])
+        expected_slice = np.array([0.5121, 0.5714, 0.4827, 0.5057, 0.5646, 0.4766, 0.5189, 0.4895, 0.4990])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

@@ -168,7 +168,7 @@ class StableDiffusion2PipelineFastTests(PipelineTesterMixin, unittest.TestCase):
        image_slice = image[0, -3:, -3:, -1]

        assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.4717, 0.5376, 0.4568, 0.5225, 0.5734, 0.4797, 0.5467, 0.5074, 0.5043])
+        expected_slice = np.array([0.4865, 0.5439, 0.4840, 0.4995, 0.5543, 0.4846, 0.5199, 0.4942, 0.5061])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

@@ -185,7 +185,7 @@ class StableDiffusion2PipelineFastTests(PipelineTesterMixin, unittest.TestCase):
        image_slice = image[0, -3:, -3:, -1]

        assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.4715, 0.5376, 0.4569, 0.5224, 0.5734, 0.4797, 0.5465, 0.5074, 0.5046])
+        expected_slice = np.array([0.4864, 0.5440, 0.4842, 0.4994, 0.5543, 0.4846, 0.5196, 0.4942, 0.5063])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

@@ -202,7 +202,7 @@ class StableDiffusion2PipelineFastTests(PipelineTesterMixin, unittest.TestCase):
        image_slice = image[0, -3:, -3:, -1]

        assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.4717, 0.5376, 0.4568, 0.5225, 0.5734, 0.4797, 0.5467, 0.5074, 0.5043])
+        expected_slice = np.array([0.4865, 0.5439, 0.4840, 0.4995, 0.5543, 0.4846, 0.5199, 0.4942, 0.5061])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

--- a/tests/pipelines/stable_diffusion_2/test_stable_diffusion_attend_and_excite.py
+++ b/tests/pipelines/stable_diffusion_2/test_stable_diffusion_attend_and_excite.py
@@ -132,9 +132,7 @@ class StableDiffusionAttendAndExcitePipelineFastTests(PipelineTesterMixin, unitt
        image_slice = image[0, -3:, -3:, -1]

        self.assertEqual(image.shape, (1, 64, 64, 3))
-        expected_slice = np.array(
-            [0.5644937, 0.60543084, 0.48239064, 0.5206757, 0.55623394, 0.46045133, 0.5100435, 0.48919064, 0.4759359]
-        )
+        expected_slice = np.array([0.5743, 0.6081, 0.4975, 0.5021, 0.5441, 0.4699, 0.4988, 0.4841, 0.4851])
        max_diff = np.abs(image_slice.flatten() - expected_slice).max()
        self.assertLessEqual(max_diff, 1e-3)

--- a/tests/pipelines/stable_diffusion_2/test_stable_diffusion_depth.py
+++ b/tests/pipelines/stable_diffusion_2/test_stable_diffusion_depth.py
@@ -289,7 +289,7 @@ class StableDiffusionDepth2ImgPipelineFastTests(PipelineTesterMixin, unittest.Te
        if torch_device == "mps":
            expected_slice = np.array([0.6071, 0.5035, 0.4378, 0.5776, 0.5753, 0.4316, 0.4513, 0.5263, 0.4546])
        else:
-            expected_slice = np.array([0.6312, 0.4984, 0.4154, 0.4788, 0.5535, 0.4599, 0.4017, 0.5359, 0.4716])
+            expected_slice = np.array([0.5435, 0.4992, 0.3783, 0.4411, 0.5842, 0.4654, 0.3786, 0.5077, 0.4655])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-3

@@ -308,9 +308,9 @@ class StableDiffusionDepth2ImgPipelineFastTests(PipelineTesterMixin, unittest.Te

        assert image.shape == (1, 32, 32, 3)
        if torch_device == "mps":
-            expected_slice = np.array([0.5825, 0.5135, 0.4095, 0.5452, 0.6059, 0.4211, 0.3994, 0.5177, 0.4335])
-        else:
            expected_slice = np.array([0.6296, 0.5125, 0.3890, 0.4456, 0.5955, 0.4621, 0.3810, 0.5310, 0.4626])
+        else:
+            expected_slice = np.array([0.6012, 0.4507, 0.3769, 0.4121, 0.5566, 0.4585, 0.3803, 0.5045, 0.4631])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-3

@@ -332,7 +332,7 @@ class StableDiffusionDepth2ImgPipelineFastTests(PipelineTesterMixin, unittest.Te
        if torch_device == "mps":
            expected_slice = np.array([0.6501, 0.5150, 0.4939, 0.6688, 0.5437, 0.5758, 0.5115, 0.4406, 0.4551])
        else:
-            expected_slice = np.array([0.6267, 0.5232, 0.6001, 0.6738, 0.5029, 0.6429, 0.5364, 0.4159, 0.4674])
+            expected_slice = np.array([0.6557, 0.6214, 0.6254, 0.5775, 0.4785, 0.5949, 0.5904, 0.4785, 0.4730])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-3

@@ -351,7 +351,7 @@ class StableDiffusionDepth2ImgPipelineFastTests(PipelineTesterMixin, unittest.Te
        if torch_device == "mps":
            expected_slice = np.array([0.53232, 0.47015, 0.40868, 0.45651, 0.4891, 0.4668, 0.4287, 0.48822, 0.47439])
        else:
-            expected_slice = np.array([0.6312, 0.4984, 0.4154, 0.4788, 0.5535, 0.4599, 0.4017, 0.5359, 0.4716])
+            expected_slice = np.array([0.5435, 0.4992, 0.3783, 0.4411, 0.5842, 0.4654, 0.3786, 0.5077, 0.4655])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-3

@@ -397,7 +397,7 @@ class StableDiffusionDepth2ImgPipelineSlowTests(unittest.TestCase):
        image_slice = image[0, 253:256, 253:256, -1].flatten()

        assert image.shape == (1, 480, 640, 3)
-        expected_slice = np.array([0.9057, 0.9365, 0.9258, 0.8937, 0.8555, 0.8541, 0.8260, 0.7747, 0.7421])
+        expected_slice = np.array([0.5435, 0.4992, 0.3783, 0.4411, 0.5842, 0.4654, 0.3786, 0.5077, 0.4655])

        assert np.abs(expected_slice - image_slice).max() < 1e-4

--- a/tests/pipelines/stable_diffusion_2/test_stable_diffusion_upscale.py
+++ b/tests/pipelines/stable_diffusion_2/test_stable_diffusion_upscale.py
@@ -154,7 +154,7 @@ class StableDiffusionUpscalePipelineFastTests(unittest.TestCase):

        expected_height_width = low_res_image.size[0] * 4
        assert image.shape == (1, expected_height_width, expected_height_width, 3)
-        expected_slice = np.array([0.2562, 0.3606, 0.4204, 0.4469, 0.4822, 0.4647, 0.5315, 0.5748, 0.5606])
+        expected_slice = np.array([0.3113, 0.3910, 0.4272, 0.4859, 0.5061, 0.4652, 0.5362, 0.5715, 0.5661])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
        assert np.abs(image_from_tuple_slice.flatten() - expected_slice).max() < 1e-2
--- a/tests/pipelines/stable_diffusion_2/test_stable_diffusion_v_pred.py
+++ b/tests/pipelines/stable_diffusion_2/test_stable_diffusion_v_pred.py
@@ -144,7 +144,7 @@ class StableDiffusion2VPredictionPipelineFastTests(unittest.TestCase):
        image_from_tuple_slice = image_from_tuple[0, -3:, -3:, -1]

        assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.6424, 0.6109, 0.494, 0.5088, 0.4984, 0.4525, 0.5059, 0.5068, 0.4474])
+        expected_slice = np.array([0.6569, 0.6525, 0.5142, 0.4968, 0.4923, 0.4601, 0.4996, 0.5041, 0.4544])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
        assert np.abs(image_from_tuple_slice.flatten() - expected_slice).max() < 1e-2
@@ -193,7 +193,7 @@ class StableDiffusion2VPredictionPipelineFastTests(unittest.TestCase):
        image_from_tuple_slice = image_from_tuple[0, -3:, -3:, -1]

        assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.4616, 0.5184, 0.4887, 0.5111, 0.4839, 0.48, 0.5119, 0.5263, 0.4776])
+        expected_slice = np.array([0.5644, 0.6514, 0.5190, 0.5663, 0.5287, 0.4953, 0.5430, 0.5243, 0.4778])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
        assert np.abs(image_from_tuple_slice.flatten() - expected_slice).max() < 1e-2
--- a/tests/pipelines/stable_diffusion_safe/test_safe_diffusion.py
+++ b/tests/pipelines/stable_diffusion_safe/test_safe_diffusion.py
@@ -154,7 +154,7 @@ class SafeDiffusionPipelineFastTests(unittest.TestCase):
        image_from_tuple_slice = image_from_tuple[0, -3:, -3:, -1]

        assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.5644, 0.6018, 0.4799, 0.5267, 0.5585, 0.4641, 0.516, 0.4964, 0.4792])
+        expected_slice = np.array([0.5756, 0.6118, 0.5005, 0.5041, 0.5471, 0.4726, 0.4976, 0.4865, 0.4864])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
        assert np.abs(image_from_tuple_slice.flatten() - expected_slice).max() < 1e-2
@@ -200,7 +200,7 @@ class SafeDiffusionPipelineFastTests(unittest.TestCase):
        image_from_tuple_slice = image_from_tuple[0, -3:, -3:, -1]

        assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.5095, 0.5674, 0.4668, 0.5126, 0.5697, 0.4675, 0.5278, 0.4964, 0.4945])
+        expected_slice = np.array([0.5125, 0.5716, 0.4828, 0.5060, 0.5650, 0.4768, 0.5185, 0.4895, 0.4993])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
        assert np.abs(image_from_tuple_slice.flatten() - expected_slice).max() < 1e-2
--- a/tests/pipelines/stable_unclip/test_stable_unclip_img2img.py
+++ b/tests/pipelines/stable_unclip/test_stable_unclip_img2img.py
@@ -47,6 +47,7 @@ class StableUnCLIPImg2ImgPipelineFastTests(PipelineTesterMixin, unittest.TestCas

        feature_extractor = CLIPImageProcessor(crop_size=32, size=32)

+        torch.manual_seed(0)
        image_encoder = CLIPVisionModelWithProjection(
            CLIPVisionConfig(
                hidden_size=embedder_hidden_size,
@@ -119,16 +120,16 @@ class StableUnCLIPImg2ImgPipelineFastTests(PipelineTesterMixin, unittest.TestCas
        components = {
            # image encoding components
            "feature_extractor": feature_extractor,
-            "image_encoder": image_encoder,
+            "image_encoder": image_encoder.eval(),
            # image noising components
-            "image_normalizer": image_normalizer,
+            "image_normalizer": image_normalizer.eval(),
            "image_noising_scheduler": image_noising_scheduler,
            # regular denoising components
            "tokenizer": tokenizer,
-            "text_encoder": text_encoder,
-            "unet": unet,
+            "text_encoder": text_encoder.eval(),
+            "unet": unet.eval(),
            "scheduler": scheduler,
-            "vae": vae,
+            "vae": vae.eval(),
        }

        return components
@@ -169,9 +170,7 @@ class StableUnCLIPImg2ImgPipelineFastTests(PipelineTesterMixin, unittest.TestCas
        image_slice = image[0, -3:, -3:, -1]

        assert image.shape == (1, 32, 32, 3)
-        expected_slice = np.array(
-            [0.34588397, 0.7747054, 0.5453714, 0.5227859, 0.57656777, 0.6532228, 0.5177634, 0.49932978, 0.56626225]
-        )
+        expected_slice = np.array([0.3872, 0.7224, 0.5601, 0.4741, 0.6872, 0.5814, 0.4636, 0.3867, 0.5078])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-3

--- a/tests/pipelines/text_to_video/test_text_to_video.py
+++ b/tests/pipelines/text_to_video/test_text_to_video.py
@@ -135,7 +135,7 @@ class TextToVideoSDPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
        image_slice = frames[0][-3:, -3:, -1]

        assert frames[0].shape == (64, 64, 3)
-        expected_slice = np.array([166, 184, 167, 118, 102, 123, 108, 93, 114])
+        expected_slice = np.array([158.0, 160.0, 153.0, 125.0, 100.0, 121.0, 111.0, 93.0, 113.0])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

--- a/tests/pipelines/vq_diffusion/test_vq_diffusion.py
+++ b/tests/pipelines/vq_diffusion/test_vq_diffusion.py
@@ -143,7 +143,7 @@ class VQDiffusionPipelineFastTests(unittest.TestCase):

        assert image.shape == (1, 24, 24, 3)

-        expected_slice = np.array([0.6583, 0.6410, 0.5325, 0.5635, 0.5563, 0.4234, 0.6008, 0.5491, 0.4880])
+        expected_slice = np.array([0.6551, 0.6168, 0.5008, 0.5676, 0.5659, 0.4295, 0.6073, 0.5599, 0.4992])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
        assert np.abs(image_from_tuple_slice.flatten() - expected_slice).max() < 1e-2
@@ -187,7 +187,7 @@ class VQDiffusionPipelineFastTests(unittest.TestCase):

        assert image.shape == (1, 24, 24, 3)

-        expected_slice = np.array([0.6647, 0.6531, 0.5303, 0.5891, 0.5726, 0.4439, 0.6304, 0.5564, 0.4912])
+        expected_slice = np.array([0.6693, 0.6075, 0.4959, 0.5701, 0.5583, 0.4333, 0.6171, 0.5684, 0.4988])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
        assert np.abs(image_from_tuple_slice.flatten() - expected_slice).max() < 1e-2
--- a/tests/test_layers_utils.py
+++ b/tests/test_layers_utils.py
@@ -411,10 +411,7 @@ class Transformer2DModelTests(unittest.TestCase):

        assert attention_scores.shape == (1, 64, 64, 64)
        output_slice = attention_scores[0, -1, -3:, -3:]
-
-        expected_slice = torch.tensor(
-            [-0.2555, -0.8877, -2.4739, -2.2251, 1.2714, 0.0807, -0.4161, -1.6408, -0.0471], device=torch_device
-        )
+        expected_slice = torch.tensor([0.0143, -0.6909, -2.1547, -1.8893, 1.4097, 0.1359, -0.2521, -1.3359, 0.2598])
        assert torch.allclose(output_slice.flatten(), expected_slice, atol=1e-3)

    def test_spatial_transformer_timestep(self):
@@ -445,14 +442,12 @@ class Transformer2DModelTests(unittest.TestCase):
        output_slice_1 = attention_scores_1[0, -1, -3:, -3:]
        output_slice_2 = attention_scores_2[0, -1, -3:, -3:]

-        expected_slice_1 = torch.tensor(
-            [-0.1874, -0.9704, -1.4290, -1.3357, 1.5138, 0.3036, -0.0976, -1.1667, 0.1283], device=torch_device
-        )
+        expected_slice = torch.tensor([-0.3923, -1.0923, -1.7144, -1.5570, 1.4154, 0.1738, -0.1157, -1.2998, -0.1703])
        expected_slice_2 = torch.tensor(
-            [-0.3493, -1.0924, -1.6161, -1.5016, 1.4245, 0.1367, -0.2526, -1.3109, -0.0547], device=torch_device
+            [-0.4311, -1.1376, -1.7732, -1.5997, 1.3450, 0.0964, -0.1569, -1.3590, -0.2348]
        )

-        assert torch.allclose(output_slice_1.flatten(), expected_slice_1, atol=1e-3)
+        assert torch.allclose(output_slice_1.flatten(), expected_slice, atol=1e-3)
        assert torch.allclose(output_slice_2.flatten(), expected_slice_2, atol=1e-3)

    def test_spatial_transformer_dropout(self):
--- a/tests/test_unet_2d_blocks.py
+++ b/tests/test_unet_2d_blocks.py
@@ -57,7 +57,7 @@ class CrossAttnDownBlock2DTests(UNetBlockTesterMixin, unittest.TestCase):
        return init_dict, inputs_dict

    def test_output(self):
-        expected_slice = [0.2440, -0.6953, -0.2140, -0.3874, 0.1966, 1.2077, 0.0441, -0.7718, 0.2800]
+        expected_slice = [0.2238, -0.7396, -0.2255, -0.3829, 0.1925, 1.1665, 0.0603, -0.7295, 0.1983]
        super().test_output(expected_slice)


@@ -175,7 +175,7 @@ class UNetMidBlock2DCrossAttnTests(UNetBlockTesterMixin, unittest.TestCase):
        return init_dict, inputs_dict

    def test_output(self):
-        expected_slice = [0.1879, 2.2653, 0.5987, 1.1568, -0.8454, -1.6109, -0.8919, 0.8306, 1.6758]
+        expected_slice = [0.0187, 2.4220, 0.4484, 1.1203, -0.6121, -1.5122, -0.8270, 0.7851, 1.8335]
        super().test_output(expected_slice)


@@ -237,7 +237,7 @@ class CrossAttnUpBlock2DTests(UNetBlockTesterMixin, unittest.TestCase):
        return init_dict, inputs_dict

    def test_output(self):
-        expected_slice = [-0.2796, -0.4364, -0.1067, -0.2693, 0.1894, 0.3869, -0.3470, 0.4584, 0.5091]
+        expected_slice = [-0.1403, -0.3515, -0.0420, -0.1425, 0.3167, 0.5094, -0.2181, 0.5931, 0.5582]
        super().test_output(expected_slice)