Qwen-Image-Edit-2509 Extra Input Problem

[Mrzhans]

我看到Qwen-image-edit-2509的报告上面支持输入controlnet类型的图达到控制的效果，那我是否可以在您提供的lora脚本上面，把extra_input设置为我们提供的mask以达到训练controlnet+lora的效果？

accelerate launch examples/qwen_image/model_training/train.py \ --dataset_base_path data/example_image_dataset \ --dataset_metadata_path data/example_image_dataset/metadata_qwen_imgae_edit_multi.json \ --data_file_keys "image,mask_image,edit_image" \ --extra_inputs "mask_image,edit_image" \ --max_pixels 1048576 \ --dataset_repeat 50 \ --model_id_with_origin_paths "Qwen/Qwen-Image-Edit-2509:transformer/diffusion_pytorch_model*.safetensors,Qwen/Qwen-Image:text_encoder/model*.safetensors,Qwen/Qwen-Image:vae/diffusion_pytorch_model.safetensors" \ --learning_rate 1e-4 \ --num_epochs 5 \ --remove_prefix_in_ckpt "pipe.dit." \ --output_path "./models/train/Qwen-Image-Edit-2509_mask_lora" \ --lora_base_model "dit" \ --lora_target_modules "to_q,to_k,to_v,add_q_proj,add_k_proj,add_v_proj,to_out.0,to_add_out,img_mlp.net.2,img_mod.1,txt_mlp.net.2,txt_mod.1" \ --lora_rank 32 \ --use_gradient_checkpointing \ --dataset_num_workers 8 \ --find_unused_parameters

[Artiprocher]

@Mrzhans 可以

[Mrzhans]

@Artiprocher 感谢回复，我发现双卡微调qwen-image-edit-2509的时候跑的非常慢。一个epoch要跑12个小时，有什么办法加速吗？下面是我的训练脚本

accelerate launch examples/qwen_image/model_training/train.py \
  --dataset_base_path data/qwen_datasets \
  --dataset_metadata_path data/metadata.json \
  --data_file_keys "image,edit_image" \
  --extra_inputs "edit_image" \
  --height 1024 \
  --width 1024 \
  --max_pixels 1048576 \
  --dataset_repeat 1 \
  --model_id_with_origin_paths "Qwen/Qwen-Image-Edit-2509:transformer/diffusion_pytorch_model*.safetensors,Qwen/Qwen-Image:text_encoder/model*.safetensors,Qwen/Qwen-Image:vae/diffusion_pytorch_model.safetensors" \
  --learning_rate 1e-4 \
  --num_epochs 100 \
  --save_steps 100 \
  --remove_prefix_in_ckpt "pipe.dit." \
  --output_path "./models/train/Qwen-Image-Edit-2509_lora" \
  --lora_base_model "dit" \
  --lora_target_modules "to_q,to_k,to_v,add_q_proj,add_k_proj,add_v_proj,to_out.0,to_add_out,img_mlp.net.2,img_mod.1,txt_mlp.net.2,txt_mod.1" \
  --lora_rank 32 \
  --use_gradient_checkpointing \
  --dataset_num_workers 8 \
  --find_unused_parameters