revert: inference_mode to no_grad

invokeai/app/invocations/compel.py

@@ -56,7 +56,7 @@ class CompelInvocation(BaseInvocation):
             },
         }
 
-    @torch.inference_mode()
+    @torch.no_grad()
     def invoke(self, context: InvocationContext) -> CompelOutput:
         tokenizer_info = context.services.model_manager.get_model(
             **self.clip.tokenizer.dict(),

invokeai/app/invocations/latent.py

@@ -285,7 +285,7 @@ class TextToLatentsInvocation(BaseInvocation):
                 # MultiControlNetModel has been refactored out, just need list[ControlNetData]
         return control_data
 
-    @torch.inference_mode()
+    @torch.no_grad()
     def invoke(self, context: InvocationContext) -> LatentsOutput:
         noise = context.services.latents.get(self.noise.latents_name)
 
@@ -369,7 +369,7 @@ class LatentsToLatentsInvocation(TextToLatentsInvocation):
             },
         }
 
-    @torch.inference_mode()
+    @torch.no_grad()
     def invoke(self, context: InvocationContext) -> LatentsOutput:
         noise = context.services.latents.get(self.noise.latents_name)
         latent = context.services.latents.get(self.latents.latents_name)
@@ -461,7 +461,7 @@ class LatentsToImageInvocation(BaseInvocation):
             },
         }
 
-    @torch.inference_mode()
+    @torch.no_grad()
     def invoke(self, context: InvocationContext) -> ImageOutput:
         latents = context.services.latents.get(self.latents.latents_name)
 
@@ -599,7 +599,7 @@ class ImageToLatentsInvocation(BaseInvocation):
             },
         }
 
-    @torch.inference_mode()
+    @torch.no_grad()
     def invoke(self, context: InvocationContext) -> LatentsOutput:
         # image = context.services.images.get(
         #     self.image.image_type, self.image.image_name

invokeai/backend/model_management/lora.py

@@ -526,7 +526,7 @@ class ModelPatcher:
     ):
         original_weights = dict()
         try:
-            with torch.inference_mode():
+            with torch.no_grad():
                 for lora, lora_weight in loras:
                     #assert lora.device.type == "cpu"
                     for layer_key, layer in lora.layers.items():
@@ -552,7 +552,7 @@ class ModelPatcher:
             yield # wait for context manager exit
 
         finally:
-            with torch.inference_mode():
+            with torch.no_grad():
                 for module_key, weight in original_weights.items():
                     model.get_submodule(module_key).weight.copy_(weight)