mirror of
https://github.com/modelscope/DiffSynth-Studio.git
synced 2026-03-18 22:08:13 +00:00
42 lines
2.6 KiB
Bash
42 lines
2.6 KiB
Bash
modelscope download --dataset DiffSynth-Studio/diffsynth_example_dataset --include "mova/MOVA-720P-I2AV/*" --local_dir ./data/diffsynth_example_dataset
|
|
|
|
accelerate launch --config_file examples/wanvideo/model_training/full/accelerate_config_14B.yaml examples/mova/model_training/train.py \
|
|
--dataset_base_path data/diffsynth_example_dataset/mova/MOVA-720P-I2AV \
|
|
--dataset_metadata_path data/diffsynth_example_dataset/mova/MOVA-720P-I2AV/metadata.csv \
|
|
--data_file_keys "video,input_audio" \
|
|
--extra_inputs "input_audio,input_image" \
|
|
--height 720 \
|
|
--width 1280 \
|
|
--num_frames 121 \
|
|
--dataset_repeat 100 \
|
|
--model_id_with_origin_paths "openmoss/MOVA-720p:video_dit/diffusion_pytorch_model-*.safetensors,openmoss/MOVA-720p:audio_dit/diffusion_pytorch_model.safetensors,openmoss/MOVA-720p:dual_tower_bridge/diffusion_pytorch_model.safetensors,openmoss/MOVA-720p:audio_vae/diffusion_pytorch_model.safetensors,DiffSynth-Studio/Wan-Series-Converted-Safetensors:Wan2.1_VAE.safetensors,DiffSynth-Studio/Wan-Series-Converted-Safetensors:models_t5_umt5-xxl-enc-bf16.safetensors" \
|
|
--learning_rate 1e-4 \
|
|
--num_epochs 5 \
|
|
--remove_prefix_in_ckpt "pipe.video_dit." \
|
|
--output_path "./models/train/MOVA-720p-I2AV_high_noise_full" \
|
|
--trainable_models "dit" \
|
|
--max_timestep_boundary 0.358 \
|
|
--min_timestep_boundary 0 \
|
|
--use_gradient_checkpointing
|
|
# boundary corresponds to timesteps [900, 1000]
|
|
|
|
accelerate launch --config_file examples/wanvideo/model_training/full/accelerate_config_14B.yaml examples/mova/model_training/train.py \
|
|
--dataset_base_path data/diffsynth_example_dataset/mova/MOVA-720P-I2AV \
|
|
--dataset_metadata_path data/diffsynth_example_dataset/mova/MOVA-720P-I2AV/metadata.csv \
|
|
--data_file_keys "video,input_audio" \
|
|
--extra_inputs "input_audio,input_image" \
|
|
--height 720 \
|
|
--width 1280 \
|
|
--num_frames 121 \
|
|
--dataset_repeat 100 \
|
|
--model_id_with_origin_paths "openmoss/MOVA-720p:video_dit_2/diffusion_pytorch_model-*.safetensors,openmoss/MOVA-720p:audio_dit/diffusion_pytorch_model.safetensors,openmoss/MOVA-720p:dual_tower_bridge/diffusion_pytorch_model.safetensors,openmoss/MOVA-720p:audio_vae/diffusion_pytorch_model.safetensors,DiffSynth-Studio/Wan-Series-Converted-Safetensors:Wan2.1_VAE.safetensors,DiffSynth-Studio/Wan-Series-Converted-Safetensors:models_t5_umt5-xxl-enc-bf16.safetensors" \
|
|
--learning_rate 1e-4 \
|
|
--num_epochs 5 \
|
|
--remove_prefix_in_ckpt "pipe.video_dit." \
|
|
--output_path "./models/train/MOVA-720p-I2AV_low_noise_full" \
|
|
--trainable_models "dit" \
|
|
--max_timestep_boundary 1 \
|
|
--min_timestep_boundary 0.358 \
|
|
--use_gradient_checkpointing
|
|
# boundary corresponds to timesteps [0, 900)
|