diff --git a/modules/processing.py b/modules/processing.py
index 181e966c8..79a3295b0 100644
--- a/modules/processing.py
+++ b/modules/processing.py
@@ -216,7 +216,7 @@ class StableDiffusionProcessing:
         conditioning_mask = torch.nn.functional.interpolate(conditioning_mask, size=latent_image.shape[-2:])
         conditioning_mask = conditioning_mask.expand(conditioning_image.shape[0], -1, -1, -1)
         image_conditioning = torch.cat([conditioning_mask, conditioning_image], dim=1)
-        image_conditioning = image_conditioning.to(device = shared.device, dtype = source_image.dtype)
+        image_conditioning = image_conditioning.to(device=shared.device, dtype=source_image.dtype)
         return image_conditioning
 
     def img2img_image_conditioning(self, source_image, latent_image, image_mask=None):
diff --git a/modules/sd_hijack.py b/modules/sd_hijack.py
index fe9d32040..decefeb11 100644
--- a/modules/sd_hijack.py
+++ b/modules/sd_hijack.py
@@ -181,7 +181,7 @@ class StableDiffusionModelHijack:
                 if opts.cuda_compile_mode == 'ipex':
                     import intel_extension_for_pytorch as ipex # pylint: disable=import-error, unused-import
                     m.model.training = False
-                    m.model = ipex.optimize(m.model, dtype=devices.dtype, inplace=True, weights_prepack=False) # pylint: disable=attribute-defined-outside-init
+                    m.model = ipex.optimize(m.model, dtype=devices.dtype_unet, inplace=True, weights_prepack=False) # pylint: disable=attribute-defined-outside-init
                 else:
                     import torch._dynamo # pylint: disable=unused-import,redefined-outer-name
                     log_level = logging.WARNING if opts.cuda_compile_verbose else logging.CRITICAL # pylint: disable=protected-access
@@ -308,4 +308,5 @@ ldm.models.diffusion.ddim.DDIMSampler.register_buffer = register_buffer
 ldm.models.diffusion.plms.PLMSSampler.register_buffer = register_buffer
 
 # Ensure samping from Guassian for DDPM follows types
-ldm.modules.distributions.distributions.DiagonalGaussianDistribution.sample = lambda self: self.mean.to(self.parameters.dtype) + self.std.to(self.parameters.dtype) * torch.randn(self.mean.shape, dtype=self.parameters.dtype).to(device=self.parameters.device)
+if not devices.backend == 'ipex':
+    ldm.modules.distributions.distributions.DiagonalGaussianDistribution.sample = lambda self: self.mean.to(self.parameters.dtype) + self.std.to(self.parameters.dtype) * torch.randn(self.mean.shape, dtype=self.parameters.dtype).to(device=self.parameters.device)
diff --git a/modules/sd_models.py b/modules/sd_models.py
index 98a394a2d..f8011f9e5 100644
--- a/modules/sd_models.py
+++ b/modules/sd_models.py
@@ -763,7 +763,7 @@ def load_diffuser(checkpoint_info=None, already_loaded_state_dict=None, timer=No
                 shared.log.info(f"Compiling pipeline={sd_model.__class__.__name__} shape={8 * sd_model.unet.config.sample_size} mode={shared.opts.cuda_compile_mode}")
                 if shared.opts.cuda_compile_mode == 'ipex':
                     sd_model.unet.training = False
-                    sd_model.unet = torch.xpu.optimize(sd_model.unet, dtype=devices.dtype, inplace=True, weights_prepack=False) # pylint: disable=attribute-defined-outside-init
+                    sd_model.unet = torch.xpu.optimize(sd_model.unet, dtype=devices.dtype_unet, inplace=True, weights_prepack=False) # pylint: disable=attribute-defined-outside-init
                 else:
                     import torch._dynamo # pylint: disable=unused-import,redefined-outer-name
                     log_level = logging.WARNING if shared.opts.cuda_compile_verbose else logging.CRITICAL # pylint: disable=protected-access