diff --git a/comfy/model_management.py b/comfy/model_management.py
index adda3841d..f550d1d93 100644
--- a/comfy/model_management.py
+++ b/comfy/model_management.py
@@ -331,7 +331,7 @@ class LoadedModel:
         use_more_vram = lowvram_model_memory
         if use_more_vram == 0:
             use_more_vram = 1e32
-        self.model_use_more_vram(use_more_vram)
+        self.model_use_more_vram(use_more_vram, force_patch_weights=force_patch_weights)
         real_model = self.model.model
 
         if is_intel_xpu() and not args.disable_ipex_optimize and 'ipex' in globals() and real_model is not None:
diff --git a/execution.py b/execution.py
index 2677fb18d..768e35abc 100644
--- a/execution.py
+++ b/execution.py
@@ -480,6 +480,7 @@ class PromptExecutor:
                 if self.caches.outputs.get(node_id) is not None:
                     cached_nodes.append(node_id)
 
+            comfy.model_management.cleanup_models_gc()
             self.add_message("execution_cached",
                           { "nodes": cached_nodes, "prompt_id": prompt_id},
                           broadcast=False)