always compute on-the-fly lora weights when offload

backend/memory_management.py

@@ -441,9 +441,8 @@ def build_module_profile(model, model_gpu_memory_when_using_cpu_swap):
 
 
 class LoadedModel:
-    def __init__(self, model, memory_required):
+    def __init__(self, model):
         self.model = model
-        self.memory_required = memory_required
         self.model_accelerated = False
         self.device = model.load_device
         self.inclusive_memory = 0
@@ -607,16 +606,16 @@ def compute_model_gpu_memory_when_using_cpu_swap(current_free_mem, inference_mem
     return int(max(0, suggestion))
 
 
-def load_models_gpu(models, memory_required=0):
+def load_models_gpu(models, memory_required=0, hard_memory_preservation=0):
     global vram_state
 
     execution_start_time = time.perf_counter()
-    extra_mem = max(minimum_inference_memory(), memory_required)
+    extra_mem = max(minimum_inference_memory(), memory_required + hard_memory_preservation)
 
     models_to_load = []
     models_already_loaded = []
     for x in models:
-        loaded_model = LoadedModel(x, memory_required=memory_required)
+        loaded_model = LoadedModel(x)
 
         if loaded_model in current_loaded_models:
             index = current_loaded_models.index(loaded_model)
@@ -664,7 +663,7 @@ def load_models_gpu(models, memory_required=0):
             previously_loaded = loaded_model.inclusive_memory
 
             current_free_mem = get_free_memory(torch_dev)
-            inference_memory = minimum_inference_memory()
+            inference_memory = minimum_inference_memory() + hard_memory_preservation
             estimated_remaining_memory = current_free_mem - model_require - inference_memory
 
             print(f"[Memory Management] Target: {loaded_model.model.model.__class__.__name__}, Free GPU: {current_free_mem / (1024 * 1024):.2f} MB, Model Require: {model_require / (1024 * 1024):.2f} MB, Previously Loaded: {previously_loaded / (1024 * 1024):.2f} MB, Inference Require: {inference_memory / (1024 * 1024):.2f} MB, Remaining: {estimated_remaining_memory / (1024 * 1024):.2f} MB, ", end="")

backend/sampling/sampling_function.py

@@ -382,7 +382,9 @@ def sampling_prepare(unet, x):
 
     memory_management.load_models_gpu(
         models=[unet] + additional_model_patchers,
-        memory_required=unet_inference_memory + additional_inference_memory)
+        memory_required=unet_inference_memory,
+        hard_memory_preservation=additional_inference_memory
+    )
 
     if unet.has_online_lora():
         utils.nested_move_to_device(unet.lora_patches, device=unet.current_device, dtype=unet.model.computation_dtype)