theluyuan/DiffSynth-Studio
1
0
Fork 0
mirror of https://github.com/modelscope/DiffSynth-Studio.git synced 2026-03-18 22:08:13 +00:00
Code Issues Packages Projects Releases Wiki Activity

add example_dataset in training scripts (#1358)

Browse Source 
* add example_dataset in training scripts

* fix example datasets
This commit is contained in:
Zhongjie Duan
2026-03-18 15:37:03 +08:00
committed by GitHub
parent 4ec4d9c20a
commit ba0626e38f
196 changed files with 911 additions and 567 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
examples/mova/model_training/full/MOVA-360P-I2AV.sh
View File

@@ -1,6 +1,8 @@
modelscope download --dataset DiffSynth-Studio/diffsynth_example_dataset --include "mova/MOVA-360P-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/example_video_dataset/ltx2 \
--dataset_metadata_path data/example_video_dataset/ltx2_t2av.csv \
--dataset_base_path data/diffsynth_example_dataset/mova/MOVA-360P-I2AV \
--dataset_metadata_path data/diffsynth_example_dataset/mova/MOVA-360P-I2AV/metadata.csv \
--data_file_keys "video,input_audio" \
--extra_inputs "input_audio,input_image" \
--height 352 \
@@ -19,8 +21,8 @@ accelerate launch --config_file examples/wanvideo/model_training/full/accelerate
# 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/example_video_dataset/ltx2 \
--dataset_metadata_path data/example_video_dataset/ltx2_t2av.csv \
--dataset_base_path data/diffsynth_example_dataset/mova/MOVA-360P-I2AV \
--dataset_metadata_path data/diffsynth_example_dataset/mova/MOVA-360P-I2AV/metadata.csv \
--data_file_keys "video,input_audio" \
--extra_inputs "input_audio,input_image" \
--height 352 \
@@ -36,4 +38,4 @@ accelerate launch --config_file examples/wanvideo/model_training/full/accelerate
--max_timestep_boundary 1 \
--min_timestep_boundary 0.358 \
--use_gradient_checkpointing
# boundary corresponds to timesteps [0, 900)
# boundary corresponds to timesteps [0, 900)

12
examples/mova/model_training/full/MOVA-720P-I2AV.sh
View File

@@ -1,6 +1,8 @@
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/example_video_dataset/ltx2 \
--dataset_metadata_path data/example_video_dataset/ltx2_t2av.csv \
--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 \
@@ -19,8 +21,8 @@ accelerate launch --config_file examples/wanvideo/model_training/full/accelerate
# 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/example_video_dataset/ltx2 \
--dataset_metadata_path data/example_video_dataset/ltx2_t2av.csv \
--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 \
@@ -36,4 +38,4 @@ accelerate launch --config_file examples/wanvideo/model_training/full/accelerate
--max_timestep_boundary 1 \
--min_timestep_boundary 0.358 \
--use_gradient_checkpointing
# boundary corresponds to timesteps [0, 900)
# boundary corresponds to timesteps [0, 900)

48
examples/mova/model_training/lora/MOVA-360P-I2AV.sh
View File

@@ -1,6 +1,8 @@
modelscope download --dataset DiffSynth-Studio/diffsynth_example_dataset --include "mova/MOVA-360P-I2AV/*" --local_dir ./data/diffsynth_example_dataset
accelerate launch examples/mova/model_training/train.py \
--dataset_base_path data/example_video_dataset/ltx2 \
--dataset_metadata_path data/example_video_dataset/ltx2_t2av.csv \
--dataset_base_path data/diffsynth_example_dataset/mova/MOVA-360P-I2AV \
--dataset_metadata_path data/diffsynth_example_dataset/mova/MOVA-360P-I2AV/metadata.csv \
--data_file_keys "video,input_audio" \
--extra_inputs "input_audio,input_image" \
--height 352 \
@@ -20,24 +22,24 @@ accelerate launch examples/mova/model_training/train.py \
--use_gradient_checkpointing
# boundary corresponds to timesteps [900, 1000]
# accelerate launch examples/mova/model_training/train.py \
# --dataset_base_path data/example_video_dataset/ltx2 \
# --dataset_metadata_path data/example_video_dataset/ltx2_t2av.csv \
# --data_file_keys "video,input_audio" \
# --extra_inputs "input_audio,input_image" \
# --height 352 \
# --width 640 \
# --num_frames 121 \
# --dataset_repeat 100 \
# --model_id_with_origin_paths "openmoss/MOVA-360p:video_dit_2/diffusion_pytorch_model-*.safetensors,openmoss/MOVA-360p:audio_dit/diffusion_pytorch_model.safetensors,openmoss/MOVA-360p: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-360p-I2AV_low_noise_lora" \
# --lora_base_model "video_dit" \
# --lora_target_modules "q,k,v,o,ffn.0,ffn.2" \
# --lora_rank 32 \
# --max_timestep_boundary 1 \
# --min_timestep_boundary 0.358 \
# --use_gradient_checkpointing
# boundary corresponds to timesteps [0, 900)
accelerate launch examples/mova/model_training/train.py \
--dataset_base_path data/diffsynth_example_dataset/mova/MOVA-360P-I2AV \
--dataset_metadata_path data/diffsynth_example_dataset/mova/MOVA-360P-I2AV/metadata.csv \
--data_file_keys "video,input_audio" \
--extra_inputs "input_audio,input_image" \
--height 352 \
--width 640 \
--num_frames 121 \
--dataset_repeat 100 \
--model_id_with_origin_paths "openmoss/MOVA-360p:video_dit_2/diffusion_pytorch_model-*.safetensors,openmoss/MOVA-360p:audio_dit/diffusion_pytorch_model.safetensors,openmoss/MOVA-360p: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-360p-I2AV_low_noise_lora" \
--lora_base_model "video_dit" \
--lora_target_modules "q,k,v,o,ffn.0,ffn.2" \
--lora_rank 32 \
--max_timestep_boundary 1 \
--min_timestep_boundary 0.358 \
--use_gradient_checkpointing
# boundary corresponds to timesteps [0, 900)

12
examples/mova/model_training/lora/MOVA-720P-I2AV.sh
View File

@@ -1,6 +1,8 @@
modelscope download --dataset DiffSynth-Studio/diffsynth_example_dataset --include "mova/MOVA-720P-I2AV/*" --local_dir ./data/diffsynth_example_dataset
accelerate launch examples/mova/model_training/train.py \
--dataset_base_path data/example_video_dataset/ltx2 \
--dataset_metadata_path data/example_video_dataset/ltx2_t2av.csv \
--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 \
@@ -21,8 +23,8 @@ accelerate launch examples/mova/model_training/train.py \
# boundary corresponds to timesteps [900, 1000]
accelerate launch examples/mova/model_training/train.py \
--dataset_base_path data/example_video_dataset/ltx2 \
--dataset_metadata_path data/example_video_dataset/ltx2_t2av.csv \
--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 \
@@ -40,4 +42,4 @@ accelerate launch examples/mova/model_training/train.py \
--max_timestep_boundary 1 \
--min_timestep_boundary 0.358 \
--use_gradient_checkpointing
# boundary corresponds to timesteps [0, 900)
# boundary corresponds to timesteps [0, 900)

12
examples/mova/model_training/train.py
View File

@@ -148,11 +148,11 @@ if __name__ == "__main__":
max_pixels=args.max_pixels,
height=args.height,
width=args.width,
height_division_factor=model.pipe.height_division_factor,
width_division_factor=model.pipe.width_division_factor,
height_division_factor=16,
width_division_factor=16,
num_frames=args.num_frames,
time_division_factor=model.pipe.time_division_factor,
time_division_remainder=model.pipe.time_division_remainder,
time_division_factor=4,
time_division_remainder=1,
frame_rate=args.frame_rate,
fix_frame_rate=True,
)
@@ -166,8 +166,8 @@ if __name__ == "__main__":
"input_audio":
ToAbsolutePath(args.dataset_base_path) >> LoadAudioWithTorchaudio(
num_frames=args.num_frames,
time_division_factor=model.pipe.time_division_factor,
time_division_remainder=model.pipe.time_division_remainder,
time_division_factor=4,
time_division_remainder=1,
frame_rate=args.frame_rate,
),
"in_context_videos":