CPU offload mode check & Enable compile for IPEX

modules/processing_diffusers.py

@@ -211,6 +211,7 @@ def process_diffusers(p: StableDiffusionProcessing, seeds, prompts, negative_pro
         if shared.opts.diffusers_move_refiner and not shared.sd_refiner.has_accelerate:
             shared.log.debug('Diffusers: Moving refiner model to CPU')
             shared.sd_refiner.to(devices.cpu)
+            devices.torch_gc()
     else:
         results = output.images
 

modules/sd_models.py

@@ -655,6 +655,12 @@ def load_diffuser(checkpoint_info=None, already_loaded_state_dict=None, timer=No
         elif "Kandinsky" in sd_model.__class__.__name__:
             sd_model.scheduler.name = 'DDIM'
 
+        if (shared.opts.diffusers_model_cpu_offload or shared.cmd_opts.medvram) and (shared.opts.diffusers_seq_cpu_offload or shared.cmd_opts.lowvram):
+            shared.log.warning(f'Diffusers {op}: Model CPU offload (--medvram) and Sequential CPU offload (--lowvram) are not compatible')
+            shared.log.debug(f'Diffusers {op}: disable model CPU offload and --medvram')
+            shared.opts.diffusers_model_cpu_offload=False
+            shared.cmd_opts.medvram=False
+
         if hasattr(sd_model, "watermark"):
             sd_model.watermark = NoWatermark()
         sd_model.has_accelerate = False

modules/shared.py

@@ -384,8 +384,8 @@ options_templates.update(options_section(('cuda', "Compute Settings"), {
     "cudnn_benchmark": OptionInfo(False, "Enable full-depth cuDNN benchmark feature"),
     "cuda_allow_tf32": OptionInfo(True, "Allow TF32 math ops"),
     "cuda_allow_tf16_reduced": OptionInfo(True, "Allow TF16 reduced precision math ops"),
-    "cuda_compile": OptionInfo(False, "Enable model compile (experimental)"),
-    "cuda_compile_mode": OptionInfo("none", "Model compile mode (experimental)", gr.Radio, lambda: {"choices": ['none', 'inductor', 'reduce-overhead', 'cudagraphs', 'aot_ts_nvfuser', 'hidet', 'ipex']}),
+    "cuda_compile": OptionInfo(True if devices.backend == "ipex" else False, "Enable model compile (experimental)"),
+    "cuda_compile_mode": OptionInfo("ipex" if devices.backend == "ipex" else "none", "Model compile mode (experimental)", gr.Radio, lambda: {"choices": ['none', 'inductor', 'reduce-overhead', 'cudagraphs', 'aot_ts_nvfuser', 'hidet', 'ipex']}),
     "cuda_compile_fullgraph": OptionInfo(False, "Model compile fullgraph"),
     "cuda_compile_verbose": OptionInfo(False, "Model compile verbose mode"),
     "cuda_compile_errors": OptionInfo(True, "Model compile suppress errors"),