[Tests] reduce the model sizes in the SD fast tests (huggingface#7580)

* give it a shot.

* print.

* correct assertion.

* gather results from the rest of the tests.

* change the assertion values where needed.

* remove print statements.

Author: sayakpaul

tests/pipelines/stable_diffusion/test_stable_diffusion.py

@@ -126,6 +126,8 @@ class StableDiffusionPipelineFastTests(
     callback_cfg_params = TEXT_TO_IMAGE_CALLBACK_CFG_PARAMS
 
     def get_dummy_components(self, time_cond_proj_dim=None):
+        cross_attention_dim = 8
+
         torch.manual_seed(0)
         unet = UNet2DConditionModel(
             block_out_channels=(4, 8),
@@ -136,7 +138,7 @@ def get_dummy_components(self, time_cond_proj_dim=None):
             out_channels=4,
             down_block_types=("DownBlock2D", "CrossAttnDownBlock2D"),
             up_block_types=("CrossAttnUpBlock2D", "UpBlock2D"),
-            cross_attention_dim=32,
+            cross_attention_dim=cross_attention_dim,
             norm_num_groups=2,
         )
         scheduler = DDIMScheduler(
@@ -160,11 +162,11 @@ def get_dummy_components(self, time_cond_proj_dim=None):
         text_encoder_config = CLIPTextConfig(
             bos_token_id=0,
             eos_token_id=2,
-            hidden_size=32,
-            intermediate_size=64,
+            hidden_size=cross_attention_dim,
+            intermediate_size=16,
             layer_norm_eps=1e-05,
-            num_attention_heads=8,
-            num_hidden_layers=3,
+            num_attention_heads=2,
+            num_hidden_layers=2,
             pad_token_id=1,
             vocab_size=1000,
         )
@@ -212,7 +214,7 @@ def test_stable_diffusion_ddim(self):
         image_slice = image[0, -3:, -3:, -1]
 
         assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.3203, 0.4555, 0.4711, 0.3505, 0.3973, 0.4650, 0.5137, 0.3392, 0.4045])
+        expected_slice = np.array([0.1763, 0.4776, 0.4986, 0.2566, 0.3802, 0.4596, 0.5363, 0.3277, 0.3949])
 
         assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
 
@@ -232,7 +234,7 @@ def test_stable_diffusion_lcm(self):
         image_slice = image[0, -3:, -3:, -1]
 
         assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.3454, 0.5349, 0.5185, 0.2808, 0.4509, 0.4612, 0.4655, 0.3601, 0.4315])
+        expected_slice = np.array([0.2368, 0.4900, 0.5019, 0.2723, 0.4473, 0.4578, 0.4551, 0.3532, 0.4133])
 
         assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
 
@@ -254,7 +256,7 @@ def test_stable_diffusion_lcm_custom_timesteps(self):
         image_slice = image[0, -3:, -3:, -1]
 
         assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.3454, 0.5349, 0.5185, 0.2808, 0.4509, 0.4612, 0.4655, 0.3601, 0.4315])
+        expected_slice = np.array([0.2368, 0.4900, 0.5019, 0.2723, 0.4473, 0.4578, 0.4551, 0.3532, 0.4133])
 
         assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
 
@@ -373,12 +375,6 @@ def test_stable_diffusion_prompt_embeds_with_plain_negative_prompt_list(self):
 
         assert np.abs(image_slice_1.flatten() - image_slice_2.flatten()).max() < 1e-4
 
-    def test_ip_adapter_single(self):
-        expected_pipe_slice = None
-        if torch_device == "cpu":
-            expected_pipe_slice = np.array([0.3203, 0.4555, 0.4711, 0.3505, 0.3973, 0.4650, 0.5137, 0.3392, 0.4045])
-        return super().test_ip_adapter_single(expected_pipe_slice=expected_pipe_slice)
-
     def test_stable_diffusion_ddim_factor_8(self):
         device = "cpu"  # ensure determinism for the device-dependent torch.Generator
 
@@ -394,7 +390,7 @@ def test_stable_diffusion_ddim_factor_8(self):
         image_slice = image[0, -3:, -3:, -1]
 
         assert image.shape == (1, 136, 136, 3)
-        expected_slice = np.array([0.4346, 0.5621, 0.5016, 0.3926, 0.4533, 0.4134, 0.5625, 0.5632, 0.5265])
+        expected_slice = np.array([0.4720, 0.5426, 0.5160, 0.3961, 0.4696, 0.4296, 0.5738, 0.5888, 0.5481])
 
         assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
 
@@ -412,7 +408,7 @@ def test_stable_diffusion_pndm(self):
         image_slice = image[0, -3:, -3:, -1]
 
         assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.3411, 0.5032, 0.4704, 0.3135, 0.4323, 0.4740, 0.5150, 0.3498, 0.4022])
+        expected_slice = np.array([0.1941, 0.4748, 0.4880, 0.2222, 0.4221, 0.4545, 0.5604, 0.3488, 0.3902])
 
         assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
 
@@ -452,7 +448,7 @@ def test_stable_diffusion_k_lms(self):
         image_slice = image[0, -3:, -3:, -1]
 
         assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.3149, 0.5246, 0.4796, 0.3218, 0.4469, 0.4729, 0.5151, 0.3597, 0.3954])
+        expected_slice = np.array([0.2681, 0.4785, 0.4857, 0.2426, 0.4473, 0.4481, 0.5610, 0.3676, 0.3855])
 
         assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
 
@@ -471,7 +467,7 @@ def test_stable_diffusion_k_euler_ancestral(self):
         image_slice = image[0, -3:, -3:, -1]
 
         assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.3151, 0.5243, 0.4794, 0.3217, 0.4468, 0.4728, 0.5152, 0.3598, 0.3954])
+        expected_slice = np.array([0.2682, 0.4782, 0.4855, 0.2424, 0.4472, 0.4479, 0.5612, 0.3676, 0.3854])
 
         assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
 
@@ -490,7 +486,7 @@ def test_stable_diffusion_k_euler(self):
         image_slice = image[0, -3:, -3:, -1]
 
         assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.3149, 0.5246, 0.4796, 0.3218, 0.4469, 0.4729, 0.5151, 0.3597, 0.3954])
+        expected_slice = np.array([0.2681, 0.4785, 0.4857, 0.2426, 0.4473, 0.4481, 0.5610, 0.3676, 0.3855])
 
         assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
 
@@ -562,7 +558,7 @@ def test_stable_diffusion_negative_prompt(self):
         image_slice = image[0, -3:, -3:, -1]
 
         assert image.shape == (1, 64, 64, 3)
-        expected_slice = np.array([0.3458, 0.5120, 0.4800, 0.3116, 0.4348, 0.4802, 0.5237, 0.3467, 0.3991])
+        expected_slice = np.array([0.1907, 0.4709, 0.4858, 0.2224, 0.4223, 0.4539, 0.5606, 0.3489, 0.3900])
 
         assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2