updates

Author: sayakpaul

src/diffusers/pipelines/pag/pipeline_pag_sana.py

@@ -268,7 +268,8 @@ def encode_prompt(
         else:
             batch_size = prompt_embeds.shape[0]
 
-        self.tokenizer.padding_side = "right"
+        if getattr(self, "tokenizer", None) is not None:
+            self.tokenizer.padding_side = "right"
 
         # See Section 3.1. of the paper.
         max_length = max_sequence_length

src/diffusers/pipelines/sana/pipeline_sana.py

@@ -312,7 +312,8 @@ def encode_prompt(
         else:
             batch_size = prompt_embeds.shape[0]
 
-        self.tokenizer.padding_side = "right"
+        if getattr(self, "tokenizer", None) is not None:
+            self.tokenizer.padding_side = "right"
 
         # See Section 3.1. of the paper.
         max_length = max_sequence_length

tests/pipelines/controlnet_hunyuandit/test_controlnet_hunyuandit.py

@@ -178,6 +178,12 @@ def test_save_load_optional_components(self):
         # TODO(YiYi) need to fix later
         pass
 
+    @unittest.skip(
+        "Test not supported as `encode_prompt` is called two times separately which deivates from about 99% of the pipelines we have."
+    )
+    def test_encode_prompt_works_in_isolation(self):
+        pass
+
 
 @slow
 @require_torch_accelerator

tests/pipelines/hunyuan_dit/test_hunyuan_dit.py

@@ -298,6 +298,12 @@ def test_fused_qkv_projections(self):
             original_image_slice, image_slice_disabled, atol=1e-2, rtol=1e-2
         ), "Original outputs should match when fused QKV projections are disabled."
 
+    @unittest.skip(
+        "Test not supported as `encode_prompt` is called two times separately which deivates from about 99% of the pipelines we have."
+    )
+    def test_encode_prompt_works_in_isolation(self):
+        pass
+
 
 @slow
 @require_torch_accelerator

tests/pipelines/kolors/test_kolors_img2img.py

@@ -152,3 +152,7 @@ def test_inference_batch_single_identical(self):
 
     def test_float16_inference(self):
         super().test_float16_inference(expected_max_diff=7e-2)
+
+    @unittest.skip("Test not supported because kolors img2img doesn't take pooled embeds as inputs unline kolors t2i.")
+    def test_encode_prompt_works_in_isolation(self):
+        pass

tests/pipelines/latte/test_latte.py

@@ -279,6 +279,10 @@ def test_save_load_optional_components(self):
     def test_xformers_attention_forwardGenerator_pass(self):
         super()._test_xformers_attention_forwardGenerator_pass(test_mean_pixel_difference=False)
 
+    @unittest.skip("Test not supported because `encode_prompt()` has multiple returns.")
+    def test_encode_prompt_works_in_isolation(self):
+        pass
+
 
 @slow
 @require_torch_gpu

tests/pipelines/pag/test_pag_hunyuan_dit.py

@@ -357,9 +357,8 @@ def test_pag_applied_layers(self):
         pipe._set_pag_attn_processor(pag_applied_layers=pag_layers, do_classifier_free_guidance=False)
         assert len(pipe.pag_attn_processors) == 2
 
+    @unittest.skip(
+        "Test not supported as `encode_prompt` is called two times separately which deivates from about 99% of the pipelines we have."
+    )
     def test_encode_prompt_works_in_isolation(self):
-        extra_required_param_value_dict = {
-            "device": torch.device(torch_device).type,
-            "do_classifier_free_guidance": self.get_dummy_inputs(device=torch_device).get("guidance_scale", 1.0) > 1.0,
-        }
-        return super().test_encode_prompt_works_in_isolation(extra_required_param_value_dict)
+        pass

tests/pipelines/stable_audio/test_stable_audio.py

@@ -413,6 +413,10 @@ def test_sequential_cpu_offload_forward_pass(self):
     def test_sequential_offload_forward_pass_twice(self):
         pass
 
+    @unittest.skip("Test not supported because `rotary_embed_dim` doesn't have any sensible default.")
+    def test_encode_prompt_works_in_isolation(self):
+        pass
+
 
 @nightly
 @require_torch_gpu

tests/pipelines/stable_cascade/test_stable_cascade_decoder.py

@@ -307,6 +307,14 @@ def test_stable_cascade_decoder_single_prompt_multiple_image_embeddings_with_gui
             batch_size * prior_num_images_per_prompt * decoder_num_images_per_prompt
         )
 
+    def test_encode_prompt_works_in_isolation(self):
+        extra_required_param_value_dict = {
+            "device": torch.device(torch_device).type,
+            "batch_size": 1,
+            "do_classifier_free_guidance": self.get_dummy_inputs(device=torch_device).get("guidance_scale", 1.0) > 1.0,
+        }
+        return super().test_encode_prompt_works_in_isolation(extra_required_param_value_dict)
+
 
 @slow
 @require_torch_gpu

tests/pipelines/stable_cascade/test_stable_cascade_prior.py

@@ -275,6 +275,10 @@ def test_stable_cascade_decoder_prompt_embeds(self):
 
         assert np.abs(output_prompt.image_embeddings - output_prompt_embeds.image_embeddings).max() < 1e-5
 
+    @unittest.skip("Test not supported because dtype determination relies on text encoder.")
+    def test_encode_prompt_works_in_isolation(self):
+        pass
+
 
 @slow
 @require_torch_gpu