Merge pull request #2573 from raefu/ckpt-cache

add --ckpt-cache option for faster model switching

modules/sd_models.py

-import glob
+import collections
 import os.path
 import sys
 from collections import namedtuple
@@ -15,6 +15,7 @@ model_path = os.path.abspath(os.path.join(models_path, model_dir))
 
 CheckpointInfo = namedtuple("CheckpointInfo", ['filename', 'title', 'hash', 'model_name', 'config'])
 checkpoints_list = {}
+checkpoints_loaded = collections.OrderedDict()
 
 try:
     # this silences the annoying "Some weights of the model checkpoint were not used when initializing..." message at start.
@@ -132,41 +133,45 @@ def load_model_weights(model, checkpoint_info):
     checkpoint_file = checkpoint_info.filename
     sd_model_hash = checkpoint_info.hash
 
-    print(f"Loading weights [{sd_model_hash}] from {checkpoint_file}")
+    if checkpoint_info not in checkpoints_loaded:
+        print(f"Loading weights [{sd_model_hash}] from {checkpoint_file}")
 
-    pl_sd = torch.load(checkpoint_file, map_location=shared.weight_load_location)
+        pl_sd = torch.load(checkpoint_file, map_location=shared.weight_load_location)
+        if "global_step" in pl_sd:
+            print(f"Global Step: {pl_sd['global_step']}")
 
-    if "global_step" in pl_sd:
-        print(f"Global Step: {pl_sd['global_step']}")
+        sd = get_state_dict_from_checkpoint(pl_sd)
+        model.load_state_dict(sd, strict=False)
 
-    sd = get_state_dict_from_checkpoint(pl_sd)
+        if shared.cmd_opts.opt_channelslast:
+            model.to(memory_format=torch.channels_last)
 
-    model.load_state_dict(sd, strict=False)
+        if not shared.cmd_opts.no_half:
+            model.half()
 
-    if shared.cmd_opts.opt_channelslast:
-        model.to(memory_format=torch.channels_last)
+        devices.dtype = torch.float32 if shared.cmd_opts.no_half else torch.float16
+        devices.dtype_vae = torch.float32 if shared.cmd_opts.no_half or shared.cmd_opts.no_half_vae else torch.float16
 
-    if not shared.cmd_opts.no_half:
-        model.half()
+        vae_file = os.path.splitext(checkpoint_file)[0] + ".vae.pt"
 
-    devices.dtype = torch.float32 if shared.cmd_opts.no_half else torch.float16
-    devices.dtype_vae = torch.float32 if shared.cmd_opts.no_half or shared.cmd_opts.no_half_vae else torch.float16
+        if not os.path.exists(vae_file) and shared.cmd_opts.vae_path is not None:
+            vae_file = shared.cmd_opts.vae_path
 
-    vae_file = os.path.splitext(checkpoint_file)[0] + ".vae.pt"
+        if os.path.exists(vae_file):
+            print(f"Loading VAE weights from: {vae_file}")
+            vae_ckpt = torch.load(vae_file, map_location=shared.weight_load_location)
+            vae_dict = {k: v for k, v in vae_ckpt["state_dict"].items() if k[0:4] != "loss"}
+            model.first_stage_model.load_state_dict(vae_dict)
 
-    if not os.path.exists(vae_file) and shared.cmd_opts.vae_path is not None:
-        vae_file = shared.cmd_opts.vae_path
+        model.first_stage_model.to(devices.dtype_vae)
 
-    if os.path.exists(vae_file):
-        print(f"Loading VAE weights from: {vae_file}")
-
-        vae_ckpt = torch.load(vae_file, map_location=shared.weight_load_location)
-
-        vae_dict = {k: v for k, v in vae_ckpt["state_dict"].items() if k[0:4] != "loss"}
-
-        model.first_stage_model.load_state_dict(vae_dict)
-
-    model.first_stage_model.to(devices.dtype_vae)
+        checkpoints_loaded[checkpoint_info] = model.state_dict().copy()
+        while len(checkpoints_loaded) > shared.opts.sd_checkpoint_cache:
+            checkpoints_loaded.popitem(last=False)  # LRU
+    else:
+        print(f"Loading weights [{sd_model_hash}] from cache")
+        checkpoints_loaded.move_to_end(checkpoint_info)
+        model.load_state_dict(checkpoints_loaded[checkpoint_info])
 
     model.sd_model_hash = sd_model_hash
     model.sd_model_checkpoint = checkpoint_file
@@ -205,6 +210,7 @@ def reload_model_weights(sd_model, info=None):
         return
 
     if sd_model.sd_checkpoint_info.config != checkpoint_info.config:
+        checkpoints_loaded.clear()
         shared.sd_model = load_model()
         return shared.sd_model
 

modules/shared.py

@@ -242,6 +242,7 @@ options_templates.update(options_section(('training', "Training"), {
 
 options_templates.update(options_section(('sd', "Stable Diffusion"), {
     "sd_model_checkpoint": OptionInfo(None, "Stable Diffusion checkpoint", gr.Dropdown, lambda: {"choices": modules.sd_models.checkpoint_tiles()}, refresh=sd_models.list_models),
+    "sd_checkpoint_cache": OptionInfo(0, "Checkpoints to cache in RAM", gr.Slider, {"minimum": 0, "maximum": 10, "step": 1}),
     "sd_hypernetwork": OptionInfo("None", "Hypernetwork", gr.Dropdown, lambda: {"choices": ["None"] + [x for x in hypernetworks.keys()]}, refresh=reload_hypernetworks),
     "sd_hypernetwork_strength": OptionInfo(1.0, "Hypernetwork strength", gr.Slider, {"minimum": 0.0, "maximum": 1.0, "step": 0.001}),
     "img2img_color_correction": OptionInfo(False, "Apply color correction to img2img results to match original colors."),