File size: 848 Bytes
feb33a0 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 |
accelerate launch examples/wanvideo/model_training/train.py \
--dataset_base_path data/example_video_dataset \
--dataset_metadata_path data/example_video_dataset/metadata_vace.csv \
--data_file_keys "video,vace_video,vace_reference_image" \
--height 480 \
--width 832 \
--dataset_repeat 100 \
--model_id_with_origin_paths "iic/VACE-Wan2.1-1.3B-Preview:diffusion_pytorch_model*.safetensors,iic/VACE-Wan2.1-1.3B-Preview:models_t5_umt5-xxl-enc-bf16.pth,iic/VACE-Wan2.1-1.3B-Preview:Wan2.1_VAE.pth" \
--learning_rate 1e-4 \
--num_epochs 5 \
--remove_prefix_in_ckpt "pipe.vace." \
--output_path "./models/train/Wan2.1-VACE-1.3B-Preview_lora" \
--lora_base_model "vace" \
--lora_target_modules "q,k,v,o,ffn.0,ffn.2" \
--lora_rank 32 \
--extra_inputs "vace_video,vace_reference_image" \
--use_gradient_checkpointing_offload |