Add sdxl only arg

modules/cmd_args.py

@@ -119,3 +119,4 @@ parser.add_argument("--disable-all-extensions", action='store_true', help="preve
 parser.add_argument("--disable-extra-extensions", action='store_true', help="prevent all extensions except built-in from running regardless of any other settings", default=False)
 parser.add_argument("--skip-load-model-at-start", action='store_true', help="if load a model at web start, only take effect when --nowebui", )
 parser.add_argument("--opt-unet-fp8-storage", action='store_true', help="use fp8 for SD UNet to save vram", default=False)
+parser.add_argument("--opt-unet-fp8-storage-xl", action='store_true', help="use fp8 for SD UNet to save vram", default=False)

modules/sd_models.py

@@ -394,6 +394,9 @@ def load_model_weights(model, checkpoint_info: CheckpointInfo, state_dict, timer
         if shared.cmd_opts.opt_unet_fp8_storage:
             model.model.diffusion_model = model.model.diffusion_model.to(torch.float8_e4m3fn)
             timer.record("apply fp8 unet")
+        elif model.is_sdxl and shared.cmd_opts.opt_unet_fp8_storage_xl:
+            model.model.diffusion_model = model.model.diffusion_model.to(torch.float8_e4m3fn)
+            timer.record("apply fp8 unet for sdxl")
 
     devices.unet_needs_upcast = shared.cmd_opts.upcast_sampling and devices.dtype == torch.float16 and devices.dtype_unet == torch.float16