[ONNX] Support Euler schedulers (huggingface#1328)

Author: anton-l

src/diffusers/pipelines/stable_diffusion/pipeline_onnx_stable_diffusion.py

@@ -261,8 +261,10 @@ def __call__(
                 noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
 
             # compute the previous noisy sample x_t -> x_t-1
-            latents = self.scheduler.step(noise_pred, t, torch.from_numpy(latents), **extra_step_kwargs).prev_sample
-            latents = np.array(latents)
+            scheduler_output = self.scheduler.step(
+                torch.from_numpy(noise_pred), t, torch.from_numpy(latents), **extra_step_kwargs
+            )
+            latents = scheduler_output.prev_sample.numpy()
 
             # call the callback, if provided
             if callback is not None and i % callback_steps == 0:

src/diffusers/pipelines/stable_diffusion/pipeline_onnx_stable_diffusion_img2img.py

@@ -401,8 +401,10 @@ def __call__(
                 noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
 
             # compute the previous noisy sample x_t -> x_t-1
-            latents = self.scheduler.step(noise_pred, t, torch.from_numpy(latents), **extra_step_kwargs).prev_sample
-            latents = latents.numpy()
+            scheduler_output = self.scheduler.step(
+                torch.from_numpy(noise_pred), t, torch.from_numpy(latents), **extra_step_kwargs
+            )
+            latents = scheduler_output.prev_sample.numpy()
 
             # call the callback, if provided
             if callback is not None and i % callback_steps == 0:

src/diffusers/pipelines/stable_diffusion/pipeline_onnx_stable_diffusion_inpaint.py

@@ -424,8 +424,10 @@ def __call__(
                 noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
 
             # compute the previous noisy sample x_t -> x_t-1
-            latents = self.scheduler.step(noise_pred, t, torch.from_numpy(latents), **extra_step_kwargs).prev_sample
-            latents = latents.numpy()
+            scheduler_output = self.scheduler.step(
+                torch.from_numpy(noise_pred), t, torch.from_numpy(latents), **extra_step_kwargs
+            )
+            latents = scheduler_output.prev_sample.numpy()
 
             # call the callback, if provided
             if callback is not None and i % callback_steps == 0: