Fix some bad comment in training scripts (#3798)

* relax tolerance slightly

* correct incorrect naming

examples/dreambooth/train_dreambooth.py

@@ -1092,8 +1092,8 @@ def main(args):
             unet, optimizer, train_dataloader, lr_scheduler
         )
 
-    # For mixed precision training we cast the text_encoder and vae weights to half-precision
-    # as these models are only used for inference, keeping weights in full precision is not required.
+    # For mixed precision training we cast all non-trainable weigths (vae, non-lora text_encoder and non-lora unet) to half-precision
+    # as these weights are only used for inference, keeping weights in full precision is not required.
     weight_dtype = torch.float32
     if accelerator.mixed_precision == "fp16":
         weight_dtype = torch.float16

examples/dreambooth/train_dreambooth_lora.py

@@ -790,8 +790,8 @@ def main(args):
     text_encoder.requires_grad_(False)
     unet.requires_grad_(False)
 
-    # For mixed precision training we cast the text_encoder and vae weights to half-precision
-    # as these models are only used for inference, keeping weights in full precision is not required.
+    # For mixed precision training we cast all non-trainable weigths (vae, non-lora text_encoder and non-lora unet) to half-precision
+    # as these weights are only used for inference, keeping weights in full precision is not required.
     weight_dtype = torch.float32
     if accelerator.mixed_precision == "fp16":
         weight_dtype = torch.float16

examples/text_to_image/train_text_to_image.py

@@ -747,8 +747,8 @@ def main():
     if args.use_ema:
         ema_unet.to(accelerator.device)
 
-    # For mixed precision training we cast the text_encoder and vae weights to half-precision
-    # as these models are only used for inference, keeping weights in full precision is not required.
+    # For mixed precision training we cast all non-trainable weigths (vae, non-lora text_encoder and non-lora unet) to half-precision
+    # as these weights are only used for inference, keeping weights in full precision is not required.
     weight_dtype = torch.float32
     if accelerator.mixed_precision == "fp16":
         weight_dtype = torch.float16

examples/text_to_image/train_text_to_image_lora.py

@@ -430,8 +430,8 @@ def main():
 
     text_encoder.requires_grad_(False)
 
-    # For mixed precision training we cast the text_encoder and vae weights to half-precision
-    # as these models are only used for inference, keeping weights in full precision is not required.
+    # For mixed precision training we cast all non-trainable weigths (vae, non-lora text_encoder and non-lora unet) to half-precision
+    # as these weights are only used for inference, keeping weights in full precision is not required.
     weight_dtype = torch.float32
     if accelerator.mixed_precision == "fp16":
         weight_dtype = torch.float16

examples/textual_inversion/textual_inversion.py

@@ -752,8 +752,8 @@ def main():
         text_encoder, optimizer, train_dataloader, lr_scheduler
     )
 
-    # For mixed precision training we cast the unet and vae weights to half-precision
-    # as these models are only used for inference, keeping weights in full precision is not required.
+    # For mixed precision training we cast all non-trainable weigths (vae, non-lora text_encoder and non-lora unet) to half-precision
+    # as these weights are only used for inference, keeping weights in full precision is not required.
     weight_dtype = torch.float32
     if accelerator.mixed_precision == "fp16":
         weight_dtype = torch.float16