Merge pull request #5586 from wywywywy/ldsr-improvements

LDSR improvements - cache / optimization / opt_channelslast

extensions-builtin/LDSR/ldsr_model_arch.py

@@ -11,25 +11,41 @@
 
 from ldm.models.diffusion.ddim import DDIMSampler
 from ldm.util import instantiate_from_config, ismap
+from modules import shared, sd_hijack
 
 warnings.filterwarnings("ignore", category=UserWarning)
 
+cached_ldsr_model: torch.nn.Module = None
+
 
 # Create LDSR Class
 class LDSR:
     def load_model_from_config(self, half_attention):
-        print(f"Loading model from {self.modelPath}")
-        pl_sd = torch.load(self.modelPath, map_location="cpu")
-        sd = pl_sd["state_dict"]
-        config = OmegaConf.load(self.yamlPath)
-        config.model.target = "ldm.models.diffusion.ddpm.LatentDiffusionV1"
-        model = instantiate_from_config(config.model)
-        model.load_state_dict(sd, strict=False)
-        model.cuda()
-        if half_attention:
-            model = model.half()
-
-        model.eval()
+        global cached_ldsr_model
+
+        if shared.opts.ldsr_cached and cached_ldsr_model is not None:
+            print(f"Loading model from cache")
+            model: torch.nn.Module = cached_ldsr_model
+        else:
+            print(f"Loading model from {self.modelPath}")
+            pl_sd = torch.load(self.modelPath, map_location="cpu")
+            sd = pl_sd["state_dict"]
+            config = OmegaConf.load(self.yamlPath)
+            config.model.target = "ldm.models.diffusion.ddpm.LatentDiffusionV1"
+            model: torch.nn.Module = instantiate_from_config(config.model)
+            model.load_state_dict(sd, strict=False)
+            model = model.to(shared.device)
+            if half_attention:
+                model = model.half()
+            if shared.cmd_opts.opt_channelslast:
+                model = model.to(memory_format=torch.channels_last)
+
+            sd_hijack.model_hijack.hijack(model) # apply optimization
+            model.eval()
+
+            if shared.opts.ldsr_cached:
+                cached_ldsr_model = model
+
         return {"model": model}
 
     def __init__(self, model_path, yaml_path):
@@ -94,7 +110,8 @@ def super_resolution(self, image, steps=100, target_scale=2, half_attention=Fals
         down_sample_method = 'Lanczos'
 
         gc.collect()
-        torch.cuda.empty_cache()
+        if torch.cuda.is_available:
+            torch.cuda.empty_cache()
 
         im_og = image
         width_og, height_og = im_og.size
@@ -131,7 +148,9 @@ def super_resolution(self, image, steps=100, target_scale=2, half_attention=Fals
 
         del model
         gc.collect()
-        torch.cuda.empty_cache()
+        if torch.cuda.is_available:
+            torch.cuda.empty_cache()
+
         return a
 
 
@@ -146,7 +165,7 @@ def get_cond(selected_path):
     c = rearrange(c, '1 c h w -> 1 h w c')
     c = 2. * c - 1.
 
-    c = c.to(torch.device("cuda"))
+    c = c.to(shared.device)
     example["LR_image"] = c
     example["image"] = c_up
 

extensions-builtin/LDSR/scripts/ldsr_model.py

@@ -59,6 +59,7 @@ def on_ui_settings():
     import gradio as gr
 
     shared.opts.add_option("ldsr_steps", shared.OptionInfo(100, "LDSR processing steps. Lower = faster", gr.Slider, {"minimum": 1, "maximum": 200, "step": 1}, section=('upscaling', "Upscaling")))
+    shared.opts.add_option("ldsr_cached", shared.OptionInfo(False, "Cache LDSR model in memory", gr.Checkbox, {"interactive": True}, section=('upscaling', "Upscaling")))
 
 
 script_callbacks.on_ui_settings(on_ui_settings)