[SDXL LoRA] fix batch size lora (#4509)

fix batch size lora

examples/dreambooth/train_dreambooth_lora_sdxl.py

@@ -1103,11 +1103,11 @@ def main(args):
                         "time_ids": add_time_ids.repeat(elems_to_repeat, 1),
                         "text_embeds": unet_add_text_embeds.repeat(elems_to_repeat, 1),
                     }
-                    prompt_embeds = prompt_embeds.repeat(elems_to_repeat, 1, 1)
+                    prompt_embeds_input = prompt_embeds.repeat(elems_to_repeat, 1, 1)
                     model_pred = unet(
                         noisy_model_input,
                         timesteps,
-                        prompt_embeds,
+                        prompt_embeds_input,
                         added_cond_kwargs=unet_added_conditions,
                     ).sample
                 else:
@@ -1119,9 +1119,9 @@ def main(args):
                         text_input_ids_list=[tokens_one, tokens_two],
                     )
                     unet_added_conditions.update({"text_embeds": pooled_prompt_embeds.repeat(elems_to_repeat, 1)})
-                    prompt_embeds = prompt_embeds.repeat(elems_to_repeat, 1, 1)
+                    prompt_embeds_input = prompt_embeds.repeat(elems_to_repeat, 1, 1)
                     model_pred = unet(
-                        noisy_model_input, timesteps, prompt_embeds, added_cond_kwargs=unet_added_conditions
+                        noisy_model_input, timesteps, prompt_embeds_input, added_cond_kwargs=unet_added_conditions
                     ).sample
 
                 # Get the target for loss depending on the prediction type