Allow negative learning rate

This can be used to train away from a group of images you don't want
As this moves the model away from a point instead of towards it, the change in the model is unbounded
So, don't set it too low. -4e-7 seemed to work well.

Author: Cauldrath

sdxl_train.py

@@ -272,7 +272,7 @@ def fn_recursive_set_mem_eff(module: torch.nn.Module):
     # 学習を準備する：モデルを適切な状態にする
     if args.gradient_checkpointing:
         unet.enable_gradient_checkpointing()
-    train_unet = args.learning_rate > 0
+    train_unet = args.learning_rate != 0
     train_text_encoder1 = False
     train_text_encoder2 = False
 
@@ -284,8 +284,8 @@ def fn_recursive_set_mem_eff(module: torch.nn.Module):
             text_encoder2.gradient_checkpointing_enable()
         lr_te1 = args.learning_rate_te1 if args.learning_rate_te1 is not None else args.learning_rate  # 0 means not train
         lr_te2 = args.learning_rate_te2 if args.learning_rate_te2 is not None else args.learning_rate  # 0 means not train
-        train_text_encoder1 = lr_te1 > 0
-        train_text_encoder2 = lr_te2 > 0
+        train_text_encoder1 = lr_te1 != 0
+        train_text_encoder2 = lr_te2 != 0
 
         # caching one text encoder output is not supported
         if not train_text_encoder1: