Merge 1f6ab7dbfb into 2307ff6746

Signed-off-by: bigcat88 <bigcat88@icloud.com>

comfy/sd.py

@@ -111,7 +111,7 @@ class CLIP:
             model_management.load_models_gpu([self.patcher], force_full_load=True)
         self.layer_idx = None
         self.use_clip_schedule = False
-        logging.info("CLIP model load device: {}, offload device: {}, current: {}, dtype: {}".format(load_device, offload_device, params['device'], dtype))
+        logging.info("CLIP/text encoder model load device: {}, offload device: {}, current: {}, dtype: {}".format(load_device, offload_device, params['device'], dtype))
 
     def clone(self):
         n = CLIP(no_init=True)
@@ -898,7 +898,7 @@ def load_state_dict_guess_config(sd, output_vae=True, output_clip=True, output_c
     if output_model:
         model_patcher = comfy.model_patcher.ModelPatcher(model, load_device=load_device, offload_device=model_management.unet_offload_device())
         if inital_load_device != torch.device("cpu"):
-            logging.info("loaded straight to GPU")
+            logging.info("loaded diffusion model directly to GPU")
             model_management.load_models_gpu([model_patcher], force_full_load=True)
 
     return (model_patcher, clip, vae, clipvision)

main.py

@@ -166,6 +166,14 @@ def prompt_worker(q, server_instance):
         queue_item = q.get(timeout=timeout)
         if queue_item is not None:
             item, item_id = queue_item
+
+            if item[3].get("unload_models"):
+                # For those cases where the flag is set, to clear memory before execution
+                comfy.model_management.unload_all_models()
+                gc.collect()
+                comfy.model_management.soft_empty_cache()
+                last_gc_collect = time.perf_counter()
+
             execution_start_time = time.perf_counter()
             prompt_id = item[1]
             server_instance.last_prompt_id = prompt_id

ruff.toml

@@ -4,7 +4,8 @@ lint.ignore = ["ALL"]
 # Enable specific rules
 lint.select = [
     "S307",  # suspicious-eval-usage
-    "T201",  # print-usage
+    "S102", # exec
+    "T",  # print-usage
     "W",
     # The "F" series in Ruff stands for "Pyflakes" rules, which catch various Python syntax errors and undefined names.
     # See all rules here: https://docs.astral.sh/ruff/rules/#pyflakes-f