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

Merge pull request #806 from mi804/wan2.2_boundary

Browse Source 
fix training boundary for wan2.2 A14B
This commit is contained in:
Zhongjie Duan
2025-08-15 18:43:37 +08:00
committed by GitHub
parent 4d16bdf853 6a9d875d65
commit 7dc49bd036
5 changed files with 25 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
examples/wanvideo/model_inference/Wan2.2-I2V-A14B.py
View File

@@ -28,5 +28,6 @@ video = pipe(
negative_prompt="色调艳丽，过曝，静态，细节模糊不清，字幕，风格，作品，画作，画面，静止，整体发灰，最差质量，低质量，JPEG压缩残留，丑陋的，残缺的，多余的手指，画得不好的手部，画得不好的脸部，畸形的，毁容的，形态畸形的肢体，手指融合，静止不动的画面，杂乱的背景，三条腿，背景人很多，倒着走", negative_prompt="色调艳丽，过曝，静态，细节模糊不清，字幕，风格，作品，画作，画面，静止，整体发灰，最差质量，低质量，JPEG压缩残留，丑陋的，残缺的，多余的手指，画得不好的手部，画得不好的脸部，畸形的，毁容的，形态畸形的肢体，手指融合，静止不动的画面，杂乱的背景，三条腿，背景人很多，倒着走",
seed=0, tiled=True, seed=0, tiled=True,
input_image=input_image, input_image=input_image,
switch_DiT_boundary=0.9,
) )
save_video(video, "video1.mp4", fps=15, quality=5) save_video(video, "video1.mp4", fps=15, quality=5)

10
examples/wanvideo/model_training/full/Wan2.2-I2V-A14B.sh
View File

@@ -13,8 +13,9 @@ accelerate launch --config_file examples/wanvideo/model_training/full/accelerate
--trainable_models "dit" \ --trainable_models "dit" \
--extra_inputs "input_image" \ --extra_inputs "input_image" \
--use_gradient_checkpointing_offload \ --use_gradient_checkpointing_offload \
--max_timestep_boundary 1 \ --max_timestep_boundary 0.358 \
--min_timestep_boundary 0.875 --min_timestep_boundary 0
# boundary corresponds to timesteps [900, 1000]
accelerate launch --config_file examples/wanvideo/model_training/full/accelerate_config_14B.yaml examples/wanvideo/model_training/train.py \ accelerate launch --config_file examples/wanvideo/model_training/full/accelerate_config_14B.yaml examples/wanvideo/model_training/train.py \
--dataset_base_path data/example_video_dataset \ --dataset_base_path data/example_video_dataset \
@@ -31,5 +32,6 @@ accelerate launch --config_file examples/wanvideo/model_training/full/accelerate
--trainable_models "dit" \ --trainable_models "dit" \
--extra_inputs "input_image" \ --extra_inputs "input_image" \
--use_gradient_checkpointing_offload \ --use_gradient_checkpointing_offload \
--max_timestep_boundary 0.875 \ --max_timestep_boundary 1 \
--min_timestep_boundary 0 --min_timestep_boundary 0.358
# boundary corresponds to timesteps [0, 900)

10
examples/wanvideo/model_training/full/Wan2.2-T2V-A14B.sh
View File

@@ -11,8 +11,9 @@ accelerate launch --config_file examples/wanvideo/model_training/full/accelerate
--remove_prefix_in_ckpt "pipe.dit." \ --remove_prefix_in_ckpt "pipe.dit." \
--output_path "./models/train/Wan2.2-T2V-A14B_high_noise_full" \ --output_path "./models/train/Wan2.2-T2V-A14B_high_noise_full" \
--trainable_models "dit" \ --trainable_models "dit" \
--max_timestep_boundary 1 \ --max_timestep_boundary 0.417 \
--min_timestep_boundary 0.875 --min_timestep_boundary 0
# boundary corresponds to timesteps [875, 1000]
accelerate launch --config_file examples/wanvideo/model_training/full/accelerate_config_14B.yaml examples/wanvideo/model_training/train.py \ accelerate launch --config_file examples/wanvideo/model_training/full/accelerate_config_14B.yaml examples/wanvideo/model_training/train.py \
--dataset_base_path data/example_video_dataset \ --dataset_base_path data/example_video_dataset \
@@ -27,5 +28,6 @@ accelerate launch --config_file examples/wanvideo/model_training/full/accelerate
--remove_prefix_in_ckpt "pipe.dit." \ --remove_prefix_in_ckpt "pipe.dit." \
--output_path "./models/train/Wan2.2-T2V-A14B_low_noise_full" \ --output_path "./models/train/Wan2.2-T2V-A14B_low_noise_full" \
--trainable_models "dit" \ --trainable_models "dit" \
--max_timestep_boundary 0.875 \ --max_timestep_boundary 1 \
--min_timestep_boundary 0 --min_timestep_boundary 0.417
# boundary corresponds to timesteps [0, 875)

10
examples/wanvideo/model_training/lora/Wan2.2-I2V-A14B.sh
View File

@@ -14,8 +14,9 @@ accelerate launch examples/wanvideo/model_training/train.py \
--lora_target_modules "q,k,v,o,ffn.0,ffn.2" \ --lora_target_modules "q,k,v,o,ffn.0,ffn.2" \
--lora_rank 32 \ --lora_rank 32 \
--extra_inputs "input_image" \ --extra_inputs "input_image" \
--max_timestep_boundary 1 \ --max_timestep_boundary 0.358 \
--min_timestep_boundary 0.875 --min_timestep_boundary 0
# boundary corresponds to timesteps [900, 1000]
accelerate launch examples/wanvideo/model_training/train.py \ accelerate launch examples/wanvideo/model_training/train.py \
--dataset_base_path data/example_video_dataset \ --dataset_base_path data/example_video_dataset \
@@ -33,5 +34,6 @@ accelerate launch examples/wanvideo/model_training/train.py \
--lora_target_modules "q,k,v,o,ffn.0,ffn.2" \ --lora_target_modules "q,k,v,o,ffn.0,ffn.2" \
--lora_rank 32 \ --lora_rank 32 \
--extra_inputs "input_image" \ --extra_inputs "input_image" \
--max_timestep_boundary 0.875 \ --max_timestep_boundary 1 \
--min_timestep_boundary 0 --min_timestep_boundary 0.358
# boundary corresponds to timesteps [0, 900)

10
examples/wanvideo/model_training/lora/Wan2.2-T2V-A14B.sh
View File

@@ -13,8 +13,9 @@ accelerate launch examples/wanvideo/model_training/train.py \
--lora_base_model "dit" \ --lora_base_model "dit" \
--lora_target_modules "q,k,v,o,ffn.0,ffn.2" \ --lora_target_modules "q,k,v,o,ffn.0,ffn.2" \
--lora_rank 32 \ --lora_rank 32 \
--max_timestep_boundary 1 \ --max_timestep_boundary 0.417 \
--min_timestep_boundary 0.875 --min_timestep_boundary 0
# boundary corresponds to timesteps [875, 1000]
accelerate launch examples/wanvideo/model_training/train.py \ accelerate launch examples/wanvideo/model_training/train.py \
@@ -32,5 +33,6 @@ accelerate launch examples/wanvideo/model_training/train.py \
--lora_base_model "dit" \ --lora_base_model "dit" \
--lora_target_modules "q,k,v,o,ffn.0,ffn.2" \ --lora_target_modules "q,k,v,o,ffn.0,ffn.2" \
--lora_rank 32 \ --lora_rank 32 \
--max_timestep_boundary 0.875 \ --max_timestep_boundary 1 \
--min_timestep_boundary 0 --min_timestep_boundary 0.417
# boundary corresponds to timesteps [0, 875)