added target weight to targets

config/examples/train_slider.example.yml

@@ -158,9 +158,10 @@ config:
             # negative is the prompt for the negative side of the slider and works the same as positive
             # it does not necessarily work the same as a negative prompt when generating images
             negative: "dog"
-            # LoRA weight to train this target. I recommend 1.0. Just leave it, it won't work
-            # how you expect if you change it
-            multiplier: 1.0
+            # the loss for this target is multiplied by this number.
+            # if you are doing more than one target it may be good to set less important ones
+            # to a lower number like 0.1 so they dont outweigh the primary target
+            weight: 1.0
 
 # You can put any information you want here, and it will be saved in the model.
 # The below is an example, but you can put your grocery list in it if you want.

jobs/process/TrainSliderProcess.py

@@ -116,6 +116,7 @@ def __init__(self, **kwargs):
         self.positive: str = kwargs.get('positive', None)
         self.negative: str = kwargs.get('negative', None)
         self.multiplier: float = kwargs.get('multiplier', 1.0)
+        self.weight: float = kwargs.get('weight', 1.0)
 
 
 class SliderConfig:
@@ -137,6 +138,7 @@ def __init__(
             height=512,
             action=ACTION_TYPES_SLIDER.ERASE_NEGATIVE,
             multiplier=1.0,
+            weight=1.0
     ):
         self.target_class = target_class
         self.positive = positive
@@ -146,6 +148,7 @@ def __init__(
         self.height = height
         self.action: int = action
         self.multiplier = multiplier
+        self.weight = weight
 
 
 class TrainSliderProcess(BaseTrainProcess):
@@ -429,7 +432,8 @@ def run(self):
                             width=width,
                             height=height,
                             action=ACTION_TYPES_SLIDER.ERASE_NEGATIVE,
-                            multiplier=target.multiplier
+                            multiplier=target.multiplier,
+                            weight=target.weight
                         ),
                         # erase inverted
                         EncodedPromptPair(
@@ -440,7 +444,8 @@ def run(self):
                             width=width,
                             height=height,
                             action=ACTION_TYPES_SLIDER.ERASE_NEGATIVE,
-                            multiplier=target.multiplier * -1.0
+                            multiplier=target.multiplier * -1.0,
+                            weight=target.weight
                         ),
                         # enhance standard, swap pos neg
                         EncodedPromptPair(
@@ -451,7 +456,8 @@ def run(self):
                             width=width,
                             height=height,
                             action=ACTION_TYPES_SLIDER.ENHANCE_NEGATIVE,
-                            multiplier=target.multiplier
+                            multiplier=target.multiplier,
+                            weight=target.weight
                         ),
                         # enhance inverted
                         EncodedPromptPair(
@@ -462,7 +468,8 @@ def run(self):
                             width=width,
                             height=height,
                             action=ACTION_TYPES_SLIDER.ENHANCE_NEGATIVE,
-                            multiplier=target.multiplier * -1.0
+                            multiplier=target.multiplier * -1.0,
+                            weight=target.weight
                         ),
                     ]
 
@@ -494,6 +501,7 @@ def run(self):
             neutral = prompt_pair.neutral
             negative = prompt_pair.negative
             positive = prompt_pair.positive
+            weight = prompt_pair.weight
 
             # set network multiplier
             self.network.multiplier = prompt_pair.multiplier
@@ -621,7 +629,7 @@ def run(self):
             loss = loss_function(
                 target_latents,
                 offset_neutral,
-            )
+            ) * weight
 
             loss_float = loss.item()
             if self.train_config.optimizer.startswith('dadaptation'):