Delegate GPU layer allocation to llama.cpp's --fit

server.py

@@ -62,7 +62,6 @@ from modules.models import load_model, unload_model_if_idle
 from modules.models_settings import (
     get_fallback_settings,
     get_model_metadata,
-    update_gpu_layers_and_vram,
     update_model_parameters
 )
 from modules.shared import do_cmd_flags_warnings
@@ -315,20 +314,6 @@ if __name__ == "__main__":
         model_settings = get_model_metadata(shared.model_name)
         update_model_parameters(model_settings, initial=True)  # hijack the command-line arguments
 
-        # Auto-adjust GPU layers if not provided by user and it's a llama.cpp model
-        if 'gpu_layers' not in shared.provided_arguments and shared.args.loader == 'llama.cpp' and 'gpu_layers' in model_settings:
-            vram_usage, adjusted_layers = update_gpu_layers_and_vram(
-                shared.args.loader,
-                shared.model_name,
-                model_settings['gpu_layers'],
-                shared.args.ctx_size,
-                shared.args.cache_type,
-                auto_adjust=True,
-                for_ui=False
-            )
-
-            shared.args.gpu_layers = adjusted_layers
-
         # Load the model
         shared.model, shared.tokenizer = load_model(shared.model_name)
         if shared.args.lora: