Update pipeline_stable_diffusion_3_ipa.py

pipeline_stable_diffusion_3_ipa.py

@@ -1148,7 +1148,7 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
             print('Using primary image.')
             clip_image = clip_image.resize((max(clip_image.size), max(clip_image.size)))
             #clip_image_embeds_1 = self.encode_clip_image_emb(clip_image, device, dtype)
-            with torch.inference_mode():
+            with torch.no_grad():
                 clip_image_embeds_1 = self.clip_image_processor(images=clip_image, return_tensors="pt").pixel_values
                 print('clip output shape: ', clip_image_embeds_1.shape)
                 clip_image_embeds_1 = clip_image_embeds_1.to(device, dtype=dtype)
@@ -1159,7 +1159,7 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
         if clip_image_2 != None:
             print('Using secondary image.')
             clip_image_2 = clip_image_2.resize((max(clip_image_2.size), max(clip_image_2.size)))
-            with torch.inference_mode():
+            with torch.no_grad():
                 clip_image_embeds_2 = self.clip_image_processor(images=clip_image_2, return_tensors="pt").pixel_values
                 clip_image_embeds_2 = clip_image_embeds_2.to(device, dtype=dtype)
                 clip_image_embeds_2 = self.image_encoder(clip_image_embeds_2, output_hidden_states=True).hidden_states[-2]
@@ -1168,7 +1168,7 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
         if clip_image_3 != None:
             print('Using tertiary image.')
             clip_image_3 = clip_image_3.resize((max(clip_image_3.size), max(clip_image_3.size)))
-            with torch.inference_mode():
+            with torch.no_grad():
                 clip_image_embeds_3 = self.clip_image_processor(images=clip_image_3, return_tensors="pt").pixel_values
                 clip_image_embeds_3 = clip_image_embeds_3.to(device, dtype=dtype)
                 clip_image_embeds_3 = self.image_encoder(clip_image_embeds_3, output_hidden_states=True).hidden_states[-2]
@@ -1177,7 +1177,7 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
         if clip_image_4 != None:
             print('Using quaternary image.')
             clip_image_4 = clip_image_4.resize((max(clip_image_4.size), max(clip_image_4.size)))
-            with torch.inference_mode():
+            with torch.no_grad():
                 clip_image_embeds_4 = self.clip_image_processor(images=clip_image_4, return_tensors="pt").pixel_values
                 clip_image_embeds_4 = clip_image_embeds_4.to(device, dtype=dtype)
                 clip_image_embeds_4 = self.image_encoder(clip_image_embeds_4, output_hidden_states=True).hidden_states[-2]
@@ -1186,7 +1186,7 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
         if clip_image_5 != None:
             print('Using quinary image.')
             clip_image_5 = clip_image_5.resize((max(clip_image_5.size), max(clip_image_5.size)))
-            with torch.inference_mode():
+            with torch.no_grad():
                 clip_image_embeds_5 = self.clip_image_processor(images=clip_image_5, return_tensors="pt").pixel_values
                 clip_image_embeds_5 = clip_image_embeds_5.to(device, dtype=dtype)
                 clip_image_embeds_5 = self.image_encoder(clip_image_embeds_5, output_hidden_states=True).hidden_states[-2]