mirror of
https://github.com/modelscope/DiffSynth-Studio.git
synced 2026-03-18 13:58:15 +00:00
40 lines
1.9 KiB
Bash
40 lines
1.9 KiB
Bash
modelscope download --dataset DiffSynth-Studio/diffsynth_example_dataset --include "wanvideo/Wan2.2-Fun-A14B-InP/*" --local_dir ./data/diffsynth_example_dataset
|
|
|
|
accelerate launch examples/wanvideo/model_training/train.py \
|
|
--dataset_base_path data/diffsynth_example_dataset/wanvideo/Wan2.2-Fun-A14B-InP \
|
|
--dataset_metadata_path data/diffsynth_example_dataset/wanvideo/Wan2.2-Fun-A14B-InP/metadata.csv \
|
|
--height 480 \
|
|
--width 832 \
|
|
--dataset_repeat 100 \
|
|
--model_id_with_origin_paths "PAI/Wan2.2-Fun-A14B-InP:high_noise_model/diffusion_pytorch_model*.safetensors,PAI/Wan2.2-Fun-A14B-InP:models_t5_umt5-xxl-enc-bf16.pth,PAI/Wan2.2-Fun-A14B-InP:Wan2.1_VAE.pth" \
|
|
--learning_rate 1e-4 \
|
|
--num_epochs 5 \
|
|
--remove_prefix_in_ckpt "pipe.dit." \
|
|
--output_path "./models/train/Wan2.2-Fun-A14B-InP_high_noise_lora" \
|
|
--lora_base_model "dit" \
|
|
--lora_target_modules "q,k,v,o,ffn.0,ffn.2" \
|
|
--lora_rank 32 \
|
|
--extra_inputs "input_image,end_image" \
|
|
--max_timestep_boundary 0.358 \
|
|
--min_timestep_boundary 0
|
|
# boundary corresponds to timesteps [900, 1000]
|
|
|
|
accelerate launch examples/wanvideo/model_training/train.py \
|
|
--dataset_base_path data/diffsynth_example_dataset/wanvideo/Wan2.2-Fun-A14B-InP \
|
|
--dataset_metadata_path data/diffsynth_example_dataset/wanvideo/Wan2.2-Fun-A14B-InP/metadata.csv \
|
|
--height 480 \
|
|
--width 832 \
|
|
--dataset_repeat 100 \
|
|
--model_id_with_origin_paths "PAI/Wan2.2-Fun-A14B-InP:low_noise_model/diffusion_pytorch_model*.safetensors,PAI/Wan2.2-Fun-A14B-InP:models_t5_umt5-xxl-enc-bf16.pth,PAI/Wan2.2-Fun-A14B-InP:Wan2.1_VAE.pth" \
|
|
--learning_rate 1e-4 \
|
|
--num_epochs 5 \
|
|
--remove_prefix_in_ckpt "pipe.dit." \
|
|
--output_path "./models/train/Wan2.2-Fun-A14B-InP_low_noise_lora" \
|
|
--lora_base_model "dit" \
|
|
--lora_target_modules "q,k,v,o,ffn.0,ffn.2" \
|
|
--lora_rank 32 \
|
|
--extra_inputs "input_image,end_image" \
|
|
--max_timestep_boundary 1 \
|
|
--min_timestep_boundary 0.358
|
|
# boundary corresponds to timesteps [0, 900]
|