remove tests that directly leveraged encode_prompt() in some way or the other.

Author: sayakpaul

tests/pipelines/animatediff/test_animatediff_sdxl.py

@@ -248,33 +248,6 @@ def test_to_dtype(self):
         model_dtypes = [component.dtype for component in pipe.components.values() if hasattr(component, "dtype")]
         self.assertTrue(all(dtype == torch.float16 for dtype in model_dtypes))
 
-    def test_prompt_embeds(self):
-        components = self.get_dummy_components()
-        pipe = self.pipeline_class(**components)
-        pipe.set_progress_bar_config(disable=None)
-        pipe.to(torch_device)
-
-        inputs = self.get_dummy_inputs(torch_device)
-        prompt = inputs.pop("prompt")
-
-        (
-            prompt_embeds,
-            negative_prompt_embeds,
-            pooled_prompt_embeds,
-            negative_pooled_prompt_embeds,
-        ) = pipe.encode_prompt(prompt)
-
-        pipe(
-            **inputs,
-            prompt_embeds=prompt_embeds,
-            negative_prompt_embeds=negative_prompt_embeds,
-            pooled_prompt_embeds=pooled_prompt_embeds,
-            negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
-        )
-
-    def test_save_load_optional_components(self):
-        self._test_save_load_optional_components()
-
     @unittest.skipIf(
         torch_device != "cuda" or not is_xformers_available(),
         reason="XFormers attention is only available with CUDA and `xformers` installed",

tests/pipelines/aura_flow/test_pipeline_aura_flow.py

@@ -5,9 +5,6 @@
 from transformers import AutoTokenizer, UMT5EncoderModel
 
 from diffusers import AuraFlowPipeline, AuraFlowTransformer2DModel, AutoencoderKL, FlowMatchEulerDiscreteScheduler
-from diffusers.utils.testing_utils import (
-    torch_device,
-)
 
 from ..test_pipelines_common import (
     PipelineTesterMixin,
@@ -90,37 +87,6 @@ def get_dummy_inputs(self, device, seed=0):
         }
         return inputs
 
-    def test_aura_flow_prompt_embeds(self):
-        pipe = self.pipeline_class(**self.get_dummy_components()).to(torch_device)
-        inputs = self.get_dummy_inputs(torch_device)
-
-        output_with_prompt = pipe(**inputs).images[0]
-
-        inputs = self.get_dummy_inputs(torch_device)
-        prompt = inputs.pop("prompt")
-
-        do_classifier_free_guidance = inputs["guidance_scale"] > 1
-        (
-            prompt_embeds,
-            prompt_attention_mask,
-            negative_prompt_embeds,
-            negative_prompt_attention_mask,
-        ) = pipe.encode_prompt(
-            prompt,
-            do_classifier_free_guidance=do_classifier_free_guidance,
-            device=torch_device,
-        )
-        output_with_embeds = pipe(
-            prompt_embeds=prompt_embeds,
-            prompt_attention_mask=prompt_attention_mask,
-            negative_prompt_embeds=negative_prompt_embeds,
-            negative_prompt_attention_mask=negative_prompt_attention_mask,
-            **inputs,
-        ).images[0]
-
-        max_diff = np.abs(output_with_prompt - output_with_embeds).max()
-        assert max_diff < 1e-4
-
     def test_attention_slicing_forward_pass(self):
         # Attention slicing needs to implemented differently for this because how single DiT and MMDiT
         # blocks interfere with each other.

tests/pipelines/controlnet/test_controlnet_sdxl.py

@@ -296,45 +296,6 @@ def test_stable_diffusion_xl_multi_prompts(self):
         # ensure the results are not equal
         assert np.abs(image_slice_1.flatten() - image_slice_3.flatten()).max() > 1e-4
 
-    # Copied from test_stable_diffusion_xl.py
-    def test_stable_diffusion_xl_prompt_embeds(self):
-        components = self.get_dummy_components()
-        sd_pipe = self.pipeline_class(**components)
-        sd_pipe = sd_pipe.to(torch_device)
-        sd_pipe = sd_pipe.to(torch_device)
-        sd_pipe.set_progress_bar_config(disable=None)
-
-        # forward without prompt embeds
-        inputs = self.get_dummy_inputs(torch_device)
-        inputs["prompt"] = 2 * [inputs["prompt"]]
-        inputs["num_images_per_prompt"] = 2
-
-        output = sd_pipe(**inputs)
-        image_slice_1 = output.images[0, -3:, -3:, -1]
-
-        # forward with prompt embeds
-        inputs = self.get_dummy_inputs(torch_device)
-        prompt = 2 * [inputs.pop("prompt")]
-
-        (
-            prompt_embeds,
-            negative_prompt_embeds,
-            pooled_prompt_embeds,
-            negative_pooled_prompt_embeds,
-        ) = sd_pipe.encode_prompt(prompt)
-
-        output = sd_pipe(
-            **inputs,
-            prompt_embeds=prompt_embeds,
-            negative_prompt_embeds=negative_prompt_embeds,
-            pooled_prompt_embeds=pooled_prompt_embeds,
-            negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
-        )
-        image_slice_2 = output.images[0, -3:, -3:, -1]
-
-        # make sure that it's equal
-        assert np.abs(image_slice_1.flatten() - image_slice_2.flatten()).max() < 1e-4
-
     def test_controlnet_sdxl_guess(self):
         device = "cpu"
 

tests/pipelines/controlnet/test_controlnet_sdxl_img2img.py

@@ -327,42 +327,3 @@ def test_stable_diffusion_xl_multi_prompts(self):
 
         # ensure the results are not equal
         assert np.abs(image_slice_1.flatten() - image_slice_3.flatten()).max() > 1e-4
-
-    # Copied from test_stable_diffusion_xl.py
-    def test_stable_diffusion_xl_prompt_embeds(self):
-        components = self.get_dummy_components()
-        sd_pipe = self.pipeline_class(**components)
-        sd_pipe = sd_pipe.to(torch_device)
-        sd_pipe = sd_pipe.to(torch_device)
-        sd_pipe.set_progress_bar_config(disable=None)
-
-        # forward without prompt embeds
-        inputs = self.get_dummy_inputs(torch_device)
-        inputs["prompt"] = 2 * [inputs["prompt"]]
-        inputs["num_images_per_prompt"] = 2
-
-        output = sd_pipe(**inputs)
-        image_slice_1 = output.images[0, -3:, -3:, -1]
-
-        # forward with prompt embeds
-        inputs = self.get_dummy_inputs(torch_device)
-        prompt = 2 * [inputs.pop("prompt")]
-
-        (
-            prompt_embeds,
-            negative_prompt_embeds,
-            pooled_prompt_embeds,
-            negative_pooled_prompt_embeds,
-        ) = sd_pipe.encode_prompt(prompt)
-
-        output = sd_pipe(
-            **inputs,
-            prompt_embeds=prompt_embeds,
-            negative_prompt_embeds=negative_prompt_embeds,
-            pooled_prompt_embeds=pooled_prompt_embeds,
-            negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
-        )
-        image_slice_2 = output.images[0, -3:, -3:, -1]
-
-        # make sure that it's equal
-        assert np.abs(image_slice_1.flatten() - image_slice_2.flatten()).max() < 1e-4

tests/pipelines/controlnet_xs/test_controlnetxs_sdxl.py

@@ -283,49 +283,6 @@ def test_stable_diffusion_xl_multi_prompts(self):
         # ensure the results are not equal
         assert np.abs(image_slice_1.flatten() - image_slice_3.flatten()).max() > 1e-4
 
-    # Copied from test_stable_diffusion_xl.py
-    def test_stable_diffusion_xl_prompt_embeds(self):
-        components = self.get_dummy_components()
-        sd_pipe = self.pipeline_class(**components)
-        sd_pipe = sd_pipe.to(torch_device)
-        sd_pipe = sd_pipe.to(torch_device)
-        sd_pipe.set_progress_bar_config(disable=None)
-
-        # forward without prompt embeds
-        inputs = self.get_dummy_inputs(torch_device)
-        inputs["prompt"] = 2 * [inputs["prompt"]]
-        inputs["num_images_per_prompt"] = 2
-
-        output = sd_pipe(**inputs)
-        image_slice_1 = output.images[0, -3:, -3:, -1]
-
-        # forward with prompt embeds
-        inputs = self.get_dummy_inputs(torch_device)
-        prompt = 2 * [inputs.pop("prompt")]
-
-        (
-            prompt_embeds,
-            negative_prompt_embeds,
-            pooled_prompt_embeds,
-            negative_pooled_prompt_embeds,
-        ) = sd_pipe.encode_prompt(prompt)
-
-        output = sd_pipe(
-            **inputs,
-            prompt_embeds=prompt_embeds,
-            negative_prompt_embeds=negative_prompt_embeds,
-            pooled_prompt_embeds=pooled_prompt_embeds,
-            negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
-        )
-        image_slice_2 = output.images[0, -3:, -3:, -1]
-
-        # make sure that it's equal
-        assert np.abs(image_slice_1.flatten() - image_slice_2.flatten()).max() < 1.1e-4
-
-    # Copied from test_stable_diffusion_xl.py
-    def test_save_load_optional_components(self):
-        self._test_save_load_optional_components()
-
     # Copied from test_controlnetxs.py
     def test_to_dtype(self):
         components = self.get_dummy_components()

tests/pipelines/stable_diffusion/test_stable_diffusion.py

@@ -375,84 +375,6 @@ def test_stable_diffusion_negative_prompt_embeds(self):
 
         assert np.abs(image_slice_1.flatten() - image_slice_2.flatten()).max() < 1e-4
 
-    def test_stable_diffusion_prompt_embeds_no_text_encoder_or_tokenizer(self):
-        components = self.get_dummy_components()
-        sd_pipe = StableDiffusionPipeline(**components)
-        sd_pipe = sd_pipe.to(torch_device)
-        sd_pipe.set_progress_bar_config(disable=None)
-
-        inputs = self.get_dummy_inputs(torch_device)
-        inputs["negative_prompt"] = "this is a negative prompt"
-
-        # forward
-        output = sd_pipe(**inputs)
-        image_slice_1 = output.images[0, -3:, -3:, -1]
-
-        inputs = self.get_dummy_inputs(torch_device)
-        prompt = inputs.pop("prompt")
-        negative_prompt = "this is a negative prompt"
-
-        prompt_embeds, negative_prompt_embeds = sd_pipe.encode_prompt(
-            prompt,
-            torch_device,
-            1,
-            True,
-            negative_prompt=negative_prompt,
-            prompt_embeds=None,
-            negative_prompt_embeds=None,
-        )
-
-        inputs["prompt_embeds"] = prompt_embeds
-        inputs["negative_prompt_embeds"] = negative_prompt_embeds
-
-        sd_pipe.text_encoder = None
-        sd_pipe.tokenizer = None
-
-        # forward
-        output = sd_pipe(**inputs)
-        image_slice_2 = output.images[0, -3:, -3:, -1]
-
-        assert np.abs(image_slice_1.flatten() - image_slice_2.flatten()).max() < 1e-4
-
-    def test_stable_diffusion_prompt_embeds_with_plain_negative_prompt_list(self):
-        components = self.get_dummy_components()
-        sd_pipe = StableDiffusionPipeline(**components)
-        sd_pipe = sd_pipe.to(torch_device)
-        sd_pipe = sd_pipe.to(torch_device)
-        sd_pipe.set_progress_bar_config(disable=None)
-
-        inputs = self.get_dummy_inputs(torch_device)
-        negative_prompt = 3 * ["this is a negative prompt"]
-        inputs["negative_prompt"] = negative_prompt
-        inputs["prompt"] = 3 * [inputs["prompt"]]
-
-        # forward
-        output = sd_pipe(**inputs)
-        image_slice_1 = output.images[0, -3:, -3:, -1]
-
-        inputs = self.get_dummy_inputs(torch_device)
-        inputs["negative_prompt"] = negative_prompt
-        prompt = 3 * [inputs.pop("prompt")]
-
-        text_inputs = sd_pipe.tokenizer(
-            prompt,
-            padding="max_length",
-            max_length=sd_pipe.tokenizer.model_max_length,
-            truncation=True,
-            return_tensors="pt",
-        )
-        text_inputs = text_inputs["input_ids"].to(torch_device)
-
-        prompt_embeds = sd_pipe.text_encoder(text_inputs)[0]
-
-        inputs["prompt_embeds"] = prompt_embeds
-
-        # forward
-        output = sd_pipe(**inputs)
-        image_slice_2 = output.images[0, -3:, -3:, -1]
-
-        assert np.abs(image_slice_1.flatten() - image_slice_2.flatten()).max() < 1e-4
-
     def test_stable_diffusion_ddim_factor_8(self):
         device = "cpu"  # ensure determinism for the device-dependent torch.Generator
 

tests/pipelines/stable_diffusion_3/test_pipeline_stable_diffusion_3.py

@@ -156,39 +156,6 @@ def test_stable_diffusion_3_different_negative_prompts(self):
         # Outputs should be different here
         assert max_diff > 1e-2
 
-    def test_stable_diffusion_3_prompt_embeds(self):
-        pipe = self.pipeline_class(**self.get_dummy_components()).to(torch_device)
-        inputs = self.get_dummy_inputs(torch_device)
-
-        output_with_prompt = pipe(**inputs).images[0]
-
-        inputs = self.get_dummy_inputs(torch_device)
-        prompt = inputs.pop("prompt")
-
-        do_classifier_free_guidance = inputs["guidance_scale"] > 1
-        (
-            prompt_embeds,
-            negative_prompt_embeds,
-            pooled_prompt_embeds,
-            negative_pooled_prompt_embeds,
-        ) = pipe.encode_prompt(
-            prompt,
-            prompt_2=None,
-            prompt_3=None,
-            do_classifier_free_guidance=do_classifier_free_guidance,
-            device=torch_device,
-        )
-        output_with_embeds = pipe(
-            prompt_embeds=prompt_embeds,
-            negative_prompt_embeds=negative_prompt_embeds,
-            pooled_prompt_embeds=pooled_prompt_embeds,
-            negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
-            **inputs,
-        ).images[0]
-
-        max_diff = np.abs(output_with_prompt - output_with_embeds).max()
-        assert max_diff < 1e-4
-
     def test_fused_qkv_projections(self):
         device = "cpu"  # ensure determinism for the device-dependent torch.Generator
         components = self.get_dummy_components()

tests/pipelines/stable_diffusion_3/test_pipeline_stable_diffusion_3_img2img.py

@@ -159,39 +159,7 @@ def test_stable_diffusion_3_img2img_different_negative_prompts(self):
         # Outputs should be different here
         assert max_diff > 1e-2
 
-    def test_stable_diffusion_3_img2img_prompt_embeds(self):
-        pipe = self.pipeline_class(**self.get_dummy_components()).to(torch_device)
-        inputs = self.get_dummy_inputs(torch_device)
-
-        output_with_prompt = pipe(**inputs).images[0]
-
-        inputs = self.get_dummy_inputs(torch_device)
-        prompt = inputs.pop("prompt")
-
-        do_classifier_free_guidance = inputs["guidance_scale"] > 1
-        (
-            prompt_embeds,
-            negative_prompt_embeds,
-            pooled_prompt_embeds,
-            negative_pooled_prompt_embeds,
-        ) = pipe.encode_prompt(
-            prompt,
-            prompt_2=None,
-            prompt_3=None,
-            do_classifier_free_guidance=do_classifier_free_guidance,
-            device=torch_device,
-        )
-        output_with_embeds = pipe(
-            prompt_embeds=prompt_embeds,
-            negative_prompt_embeds=negative_prompt_embeds,
-            pooled_prompt_embeds=pooled_prompt_embeds,
-            negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
-            **inputs,
-        ).images[0]
-
-        max_diff = np.abs(output_with_prompt - output_with_embeds).max()
-        assert max_diff < 1e-4
-
+    @unittest.skip("Skip for now.")
     def test_multi_vae(self):
         pass