Add torch.device

modules/extras.py

@@ -311,7 +311,7 @@ def run_MEHmodelmerger(id_task, **kwargs):  # pylint: disable=unused-argument
             kwargs.pop("beta_mid_block", None)
             kwargs.pop("beta_out_blocks", None)
             kwargs.pop("beta_preset", None)
-            
+
     if kwargs["device"] == "cuda":
         sd_models.unload_model_weights()
 

modules/merging/merge.py

@@ -1,4 +1,3 @@
-import gc
 import os
 from concurrent.futures import ThreadPoolExecutor
 from contextlib import contextmanager
@@ -10,6 +9,7 @@ import torch
 from tqdm import tqdm
 
 import modules.memstats
+import modules.devices as devices
 from modules.shared import log
 from modules.merging import merge_methods
 from modules.merging.merge_utils import WeightClass
@@ -69,7 +69,7 @@ def fix_model(model: Dict) -> Dict:
     return fix_clip(model)
 
 
-def load_sd_model(model: os.PathLike | str, device: str = "cpu") -> Dict:
+def load_sd_model(model: os.PathLike | str, device: torch.device = None) -> Dict:
     if isinstance(model, str):
         model = Path(model)
 
@@ -102,7 +102,7 @@ def log_vram(txt=""):
 def load_thetas(
     models: Dict[str, os.PathLike | str],
     prune: bool,
-    device: str,
+    device: torch.device,
     precision: str,
 ) -> Dict:
     log_vram("before loading models")
@@ -111,13 +111,12 @@ def load_thetas(
     else:
         thetas = {k: load_sd_model(m, device) for k, m in models.items()}
 
-    if device == "cuda":
-        for model_key, model in thetas.items():
-            for key, block in model.items():
-                if precision == "fp16":
-                    thetas[model_key].update({key: block.to(device).half()})
-                else:
-                    thetas[model_key].update({key: block.to(device)})
+    for model_key, model in thetas.items():
+        for key, block in model.items():
+            if precision == "fp16":
+                thetas[model_key].update({key: block.to(device).half()})
+            else:
+                thetas[model_key].update({key: block.to(device)})
 
     log_vram("models loaded")
     return thetas
@@ -129,8 +128,8 @@ def merge_models(
     precision: str = "fp16",
     weights_clip: bool = False,
     re_basin: bool = False,
-    device: str = "cpu",
-    work_device: Optional[str] = None,
+    device: torch.device = None,
+    work_device: torch.device = None,
     prune: bool = False,
     threads: int = 1,
     **kwargs,
@@ -178,7 +177,7 @@ def un_prune_model(
     if prune:
         log.info("Un-pruning merged model")
         del thetas
-        gc.collect()
+        devices.torch_gc(force=True)
         log_vram("remove thetas")
         original_a = load_sd_model(models["model_a"], device)
         for key in tqdm(original_a.keys(), desc="un-prune model a"):
@@ -189,7 +188,7 @@ def un_prune_model(
                 if precision == "fp16":
                     merged.update({key: merged[key].half()})
         del original_a
-        gc.collect()
+        devices.torch_gc(force=True)
         # log_vram("remove original_a")
         original_b = load_sd_model(models["model_b"], device)
         for key in tqdm(original_b.keys(), desc="un-prune model b"):
@@ -210,8 +209,8 @@ def simple_merge(
     merge_mode: str,
     precision: str = "fp16",
     weights_clip: bool = False,
-    device: str = "cpu",
-    work_device: Optional[str] = None,
+    device: torch.device = None,
+    work_device: torch.device = None,
     threads: int = 1,
 ) -> Dict:
     futures = []
@@ -257,8 +256,8 @@ def rebasin_merge(
     precision: str = "fp16",
     weights_clip: bool = False,
     iterations: int = 1,
-    device="cpu",
-    work_device=None,
+    device: torch.device = None,
+    work_device: torch.device = None,
     threads: int = 1,
 ):
     # WARNING: not sure how this does when 3 models are involved...
@@ -345,8 +344,8 @@ def merge_key(
     merge_mode: str,
     precision: str = "fp16",
     weights_clip: bool = False,
-    device: str = "cpu",
-    work_device: Optional[str] = None,
+    device: torch.device = None,
+    work_device: torch.device = None,
 ) -> Optional[Tuple[str, Dict]]:
     if work_device is None:
         work_device = device
@@ -413,7 +412,7 @@ def get_merge_method_args(
     current_bases: Dict,
     thetas: Dict,
     key: str,
-    work_device: str,
+    work_device: torch.device,
 ) -> Dict:
     merge_method_args = {
         "a": thetas["model_a"][key].to(work_device),

modules/merging/merge_model.py

@@ -1,4 +1,3 @@
-# import logging
 import os
 from dataclasses import dataclass
 
@@ -6,17 +5,16 @@ import safetensors
 import torch
 from tensordict import TensorDict
 
-# logging.getLogger("sd_meh").addHandler(logging.NullHandler())
 
 
 @dataclass
 class SDModel:
     model_path: os.PathLike
-    device: str
+    device: torch.device
 
     def load_model(self):
         # logging.info(f"Loading: {self.model_path}")
-        if self.model_path.suffix == ".safetensors":
+        if os.path.splitext(self.model_path) == ".safetensors":
             ckpt = safetensors.torch.load_file(
                 self.model_path,
                 device=self.device,