improve performance of 3-way merge on machines with not enough ram, by only accessing two of the models at a time

1 files changed, 17 insertions, 10 deletions

diff --git a/modules/extras.py b/modules/extras.py
index 0819ed37..340a45fd 100644
--- a/modules/extras.py
+++ b/modules/extras.py
@@ -175,11 +175,14 @@ def run_pnginfo(image):
 

 

 def run_modelmerger(primary_model_name, secondary_model_name, teritary_model_name, interp_method, multiplier, save_as_half, custom_name):

-    def weighted_sum(theta0, theta1, theta2, alpha):

+    def weighted_sum(theta0, theta1, alpha):

         return ((1 - alpha) * theta0) + (alpha * theta1)

 

-    def add_difference(theta0, theta1, theta2, alpha):

-        return theta0 + (theta1 - theta2) * alpha

+    def get_difference(theta1, theta2):

+        return theta1 - theta2

+

+    def add_difference(theta0, theta1_2_diff, alpha):

+        return theta0 + (alpha * theta1_2_diff)

 

     primary_model_info = sd_models.checkpoints_list[primary_model_name]

     secondary_model_info = sd_models.checkpoints_list[secondary_model_name]

@@ -201,20 +204,24 @@ def run_modelmerger(primary_model_name, secondary_model_name, teritary_model_nam
         theta_2 = None

 

     theta_funcs = {

-        "Weighted sum": weighted_sum,

-        "Add difference": add_difference,

+        "Weighted sum": (None, weighted_sum),

+        "Add difference": (get_difference, add_difference),

     }

-    theta_func = theta_funcs[interp_method]

+    theta_func1, theta_func2 = theta_funcs[interp_method]

 

     print(f"Merging...")

 

+    if theta_func1:

+        for key in tqdm.tqdm(theta_1.keys()):

+            if 'model' in key:

+                t2 = theta_2.get(key, torch.zeros_like(theta_1[key]))

+                theta_1[key] = theta_func1(theta_1[key], t2)

+    del theta_2, teritary_model

+

     for key in tqdm.tqdm(theta_0.keys()):

         if 'model' in key and key in theta_1:

-            t2 = (theta_2 or {}).get(key)

-            if t2 is None:

-                t2 = torch.zeros_like(theta_0[key])

 

-            theta_0[key] = theta_func(theta_0[key], theta_1[key], t2, multiplier)

+            theta_0[key] = theta_func2(theta_0[key], theta_1[key], multiplier)

 

             if save_as_half:

                 theta_0[key] = theta_0[key].half()