[tests] Single scheduler in lora tests (huggingface#12315)

* single scheduler please.

* up

* up

* up

Author: sayakpaul

tests/lora/test_lora_layers_auraflow.py

@@ -43,7 +43,6 @@
 class AuraFlowLoRATests(unittest.TestCase, PeftLoraLoaderMixinTests):
     pipeline_class = AuraFlowPipeline
     scheduler_cls = FlowMatchEulerDiscreteScheduler
-    scheduler_classes = [FlowMatchEulerDiscreteScheduler]
     scheduler_kwargs = {}
 
     transformer_kwargs = {

tests/lora/test_lora_layers_cogvideox.py

@@ -21,7 +21,6 @@
 
 from diffusers import (
     AutoencoderKLCogVideoX,
-    CogVideoXDDIMScheduler,
     CogVideoXDPMScheduler,
     CogVideoXPipeline,
     CogVideoXTransformer3DModel,
@@ -44,7 +43,6 @@ class CogVideoXLoRATests(unittest.TestCase, PeftLoraLoaderMixinTests):
     pipeline_class = CogVideoXPipeline
     scheduler_cls = CogVideoXDPMScheduler
     scheduler_kwargs = {"timestep_spacing": "trailing"}
-    scheduler_classes = [CogVideoXDDIMScheduler, CogVideoXDPMScheduler]
 
     transformer_kwargs = {
         "num_attention_heads": 4,

tests/lora/test_lora_layers_cogview4.py

@@ -50,7 +50,6 @@ def from_pretrained(*args, **kwargs):
 class CogView4LoRATests(unittest.TestCase, PeftLoraLoaderMixinTests):
     pipeline_class = CogView4Pipeline
     scheduler_cls = FlowMatchEulerDiscreteScheduler
-    scheduler_classes = [FlowMatchEulerDiscreteScheduler]
     scheduler_kwargs = {}
 
     transformer_kwargs = {
@@ -124,30 +123,29 @@ def test_simple_inference_save_pretrained(self):
         """
         Tests a simple usecase where users could use saving utilities for LoRA through save_pretrained
         """
-        for scheduler_cls in self.scheduler_classes:
-            components, _, _ = self.get_dummy_components(scheduler_cls)
-            pipe = self.pipeline_class(**components)
-            pipe = pipe.to(torch_device)
-            pipe.set_progress_bar_config(disable=None)
-            _, _, inputs = self.get_dummy_inputs(with_generator=False)
+        components, _, _ = self.get_dummy_components()
+        pipe = self.pipeline_class(**components)
+        pipe = pipe.to(torch_device)
+        pipe.set_progress_bar_config(disable=None)
+        _, _, inputs = self.get_dummy_inputs(with_generator=False)
 
-            output_no_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
-            self.assertTrue(output_no_lora.shape == self.output_shape)
+        output_no_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
+        self.assertTrue(output_no_lora.shape == self.output_shape)
 
-            images_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
+        images_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
 
-            with tempfile.TemporaryDirectory() as tmpdirname:
-                pipe.save_pretrained(tmpdirname)
+        with tempfile.TemporaryDirectory() as tmpdirname:
+            pipe.save_pretrained(tmpdirname)
 
-                pipe_from_pretrained = self.pipeline_class.from_pretrained(tmpdirname)
-                pipe_from_pretrained.to(torch_device)
+            pipe_from_pretrained = self.pipeline_class.from_pretrained(tmpdirname)
+            pipe_from_pretrained.to(torch_device)
 
-            images_lora_save_pretrained = pipe_from_pretrained(**inputs, generator=torch.manual_seed(0))[0]
+        images_lora_save_pretrained = pipe_from_pretrained(**inputs, generator=torch.manual_seed(0))[0]
 
-            self.assertTrue(
-                np.allclose(images_lora, images_lora_save_pretrained, atol=1e-3, rtol=1e-3),
-                "Loading from saved checkpoints should give same results.",
-            )
+        self.assertTrue(
+            np.allclose(images_lora, images_lora_save_pretrained, atol=1e-3, rtol=1e-3),
+            "Loading from saved checkpoints should give same results.",
+        )
 
     @parameterized.expand([("block_level", True), ("leaf_level", False)])
     @require_torch_accelerator

tests/lora/test_lora_layers_flux.py

@@ -55,9 +55,8 @@
 @require_peft_backend
 class FluxLoRATests(unittest.TestCase, PeftLoraLoaderMixinTests):
     pipeline_class = FluxPipeline
-    scheduler_cls = FlowMatchEulerDiscreteScheduler()
+    scheduler_cls = FlowMatchEulerDiscreteScheduler
     scheduler_kwargs = {}
-    scheduler_classes = [FlowMatchEulerDiscreteScheduler]
     transformer_kwargs = {
         "patch_size": 1,
         "in_channels": 4,
@@ -282,9 +281,8 @@ def test_simple_inference_with_text_denoiser_multi_adapter_block_lora(self):
 
 class FluxControlLoRATests(unittest.TestCase, PeftLoraLoaderMixinTests):
     pipeline_class = FluxControlPipeline
-    scheduler_cls = FlowMatchEulerDiscreteScheduler()
+    scheduler_cls = FlowMatchEulerDiscreteScheduler
     scheduler_kwargs = {}
-    scheduler_classes = [FlowMatchEulerDiscreteScheduler]
     transformer_kwargs = {
         "patch_size": 1,
         "in_channels": 8,

tests/lora/test_lora_layers_hunyuanvideo.py

@@ -51,7 +51,6 @@
 class HunyuanVideoLoRATests(unittest.TestCase, PeftLoraLoaderMixinTests):
     pipeline_class = HunyuanVideoPipeline
     scheduler_cls = FlowMatchEulerDiscreteScheduler
-    scheduler_classes = [FlowMatchEulerDiscreteScheduler]
     scheduler_kwargs = {}
 
     transformer_kwargs = {

tests/lora/test_lora_layers_ltx_video.py

@@ -37,7 +37,6 @@
 class LTXVideoLoRATests(unittest.TestCase, PeftLoraLoaderMixinTests):
     pipeline_class = LTXPipeline
     scheduler_cls = FlowMatchEulerDiscreteScheduler
-    scheduler_classes = [FlowMatchEulerDiscreteScheduler]
     scheduler_kwargs = {}
 
     transformer_kwargs = {

tests/lora/test_lora_layers_lumina2.py

@@ -39,7 +39,6 @@
 class Lumina2LoRATests(unittest.TestCase, PeftLoraLoaderMixinTests):
     pipeline_class = Lumina2Pipeline
     scheduler_cls = FlowMatchEulerDiscreteScheduler
-    scheduler_classes = [FlowMatchEulerDiscreteScheduler]
     scheduler_kwargs = {}
 
     transformer_kwargs = {
@@ -141,33 +140,30 @@ def test_simple_inference_with_text_lora_save_load(self):
         strict=False,
     )
     def test_lora_fuse_nan(self):
-        for scheduler_cls in self.scheduler_classes:
-            components, text_lora_config, denoiser_lora_config = self.get_dummy_components(scheduler_cls)
-            pipe = self.pipeline_class(**components)
-            pipe = pipe.to(torch_device)
-            pipe.set_progress_bar_config(disable=None)
-            _, _, inputs = self.get_dummy_inputs(with_generator=False)
-
-            if "text_encoder" in self.pipeline_class._lora_loadable_modules:
-                pipe.text_encoder.add_adapter(text_lora_config, "adapter-1")
-                self.assertTrue(
-                    check_if_lora_correctly_set(pipe.text_encoder), "Lora not correctly set in text encoder"
-                )
-
-            denoiser = pipe.transformer if self.unet_kwargs is None else pipe.unet
-            denoiser.add_adapter(denoiser_lora_config, "adapter-1")
-            self.assertTrue(check_if_lora_correctly_set(denoiser), "Lora not correctly set in denoiser.")
-
-            # corrupt one LoRA weight with `inf` values
-            with torch.no_grad():
-                pipe.transformer.layers[0].attn.to_q.lora_A["adapter-1"].weight += float("inf")
-
-            # with `safe_fusing=True` we should see an Error
-            with self.assertRaises(ValueError):
-                pipe.fuse_lora(components=self.pipeline_class._lora_loadable_modules, safe_fusing=True)
-
-            # without we should not see an error, but every image will be black
-            pipe.fuse_lora(components=self.pipeline_class._lora_loadable_modules, safe_fusing=False)
-            out = pipe(**inputs)[0]
-
-            self.assertTrue(np.isnan(out).all())
+        components, text_lora_config, denoiser_lora_config = self.get_dummy_components()
+        pipe = self.pipeline_class(**components)
+        pipe = pipe.to(torch_device)
+        pipe.set_progress_bar_config(disable=None)
+        _, _, inputs = self.get_dummy_inputs(with_generator=False)
+
+        if "text_encoder" in self.pipeline_class._lora_loadable_modules:
+            pipe.text_encoder.add_adapter(text_lora_config, "adapter-1")
+            self.assertTrue(check_if_lora_correctly_set(pipe.text_encoder), "Lora not correctly set in text encoder")
+
+        denoiser = pipe.transformer if self.unet_kwargs is None else pipe.unet
+        denoiser.add_adapter(denoiser_lora_config, "adapter-1")
+        self.assertTrue(check_if_lora_correctly_set(denoiser), "Lora not correctly set in denoiser.")
+
+        # corrupt one LoRA weight with `inf` values
+        with torch.no_grad():
+            pipe.transformer.layers[0].attn.to_q.lora_A["adapter-1"].weight += float("inf")
+
+        # with `safe_fusing=True` we should see an Error
+        with self.assertRaises(ValueError):
+            pipe.fuse_lora(components=self.pipeline_class._lora_loadable_modules, safe_fusing=True)
+
+        # without we should not see an error, but every image will be black
+        pipe.fuse_lora(components=self.pipeline_class._lora_loadable_modules, safe_fusing=False)
+        out = pipe(**inputs)[0]
+
+        self.assertTrue(np.isnan(out).all())

tests/lora/test_lora_layers_mochi.py

@@ -37,7 +37,6 @@
 class MochiLoRATests(unittest.TestCase, PeftLoraLoaderMixinTests):
     pipeline_class = MochiPipeline
     scheduler_cls = FlowMatchEulerDiscreteScheduler
-    scheduler_classes = [FlowMatchEulerDiscreteScheduler]
     scheduler_kwargs = {}
 
     transformer_kwargs = {

tests/lora/test_lora_layers_qwenimage.py

@@ -37,7 +37,6 @@
 class QwenImageLoRATests(unittest.TestCase, PeftLoraLoaderMixinTests):
     pipeline_class = QwenImagePipeline
     scheduler_cls = FlowMatchEulerDiscreteScheduler
-    scheduler_classes = [FlowMatchEulerDiscreteScheduler]
     scheduler_kwargs = {}
 
     transformer_kwargs = {

tests/lora/test_lora_layers_sana.py

@@ -31,9 +31,8 @@
 @require_peft_backend
 class SanaLoRATests(unittest.TestCase, PeftLoraLoaderMixinTests):
     pipeline_class = SanaPipeline
-    scheduler_cls = FlowMatchEulerDiscreteScheduler(shift=7.0)
-    scheduler_kwargs = {}
-    scheduler_classes = [FlowMatchEulerDiscreteScheduler]
+    scheduler_cls = FlowMatchEulerDiscreteScheduler
+    scheduler_kwargs = {"shift": 7.0}
     transformer_kwargs = {
         "patch_size": 1,
         "in_channels": 4,