From a18e6233b5236ebcd91333e90f5f1d16cc5b9381 Mon Sep 17 00:00:00 2001 From: Artiprocher Date: Mon, 19 Jan 2026 17:35:08 +0800 Subject: [PATCH 1/3] updata wan-vace training scripts --- .../wanvideo/model_training/lora/Wan2.1-VACE-1.3B-Preview.sh | 5 +++-- examples/wanvideo/model_training/lora/Wan2.1-VACE-1.3B.sh | 5 +++-- examples/wanvideo/model_training/lora/Wan2.1-VACE-14B.sh | 5 +++-- 3 files changed, 9 insertions(+), 6 deletions(-) diff --git a/examples/wanvideo/model_training/lora/Wan2.1-VACE-1.3B-Preview.sh b/examples/wanvideo/model_training/lora/Wan2.1-VACE-1.3B-Preview.sh index 2bcb55b..1f25eef 100644 --- a/examples/wanvideo/model_training/lora/Wan2.1-VACE-1.3B-Preview.sh +++ b/examples/wanvideo/model_training/lora/Wan2.1-VACE-1.3B-Preview.sh @@ -6,7 +6,7 @@ accelerate launch examples/wanvideo/model_training/train.py \ --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 \ + --learning_rate 5e-5 \ --num_epochs 5 \ --remove_prefix_in_ckpt "pipe.vace." \ --output_path "./models/train/Wan2.1-VACE-1.3B-Preview_lora" \ @@ -14,4 +14,5 @@ accelerate launch examples/wanvideo/model_training/train.py \ --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 \ No newline at end of file + --use_gradient_checkpointing_offload +# The learning rate is kept consistent with the settings in the original paper \ No newline at end of file diff --git a/examples/wanvideo/model_training/lora/Wan2.1-VACE-1.3B.sh b/examples/wanvideo/model_training/lora/Wan2.1-VACE-1.3B.sh index b565078..c8b77cc 100644 --- a/examples/wanvideo/model_training/lora/Wan2.1-VACE-1.3B.sh +++ b/examples/wanvideo/model_training/lora/Wan2.1-VACE-1.3B.sh @@ -6,7 +6,7 @@ accelerate launch examples/wanvideo/model_training/train.py \ --width 832 \ --dataset_repeat 100 \ --model_id_with_origin_paths "Wan-AI/Wan2.1-VACE-1.3B:diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-VACE-1.3B:models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-VACE-1.3B:Wan2.1_VAE.pth" \ - --learning_rate 1e-4 \ + --learning_rate 5e-5 \ --num_epochs 5 \ --remove_prefix_in_ckpt "pipe.vace." \ --output_path "./models/train/Wan2.1-VACE-1.3B_lora" \ @@ -14,4 +14,5 @@ accelerate launch examples/wanvideo/model_training/train.py \ --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 \ No newline at end of file + --use_gradient_checkpointing_offload +# The learning rate is kept consistent with the settings in the original paper \ No newline at end of file diff --git a/examples/wanvideo/model_training/lora/Wan2.1-VACE-14B.sh b/examples/wanvideo/model_training/lora/Wan2.1-VACE-14B.sh index 633ea0e..28bd05c 100644 --- a/examples/wanvideo/model_training/lora/Wan2.1-VACE-14B.sh +++ b/examples/wanvideo/model_training/lora/Wan2.1-VACE-14B.sh @@ -7,7 +7,7 @@ accelerate launch examples/wanvideo/model_training/train.py \ --num_frames 17 \ --dataset_repeat 100 \ --model_id_with_origin_paths "Wan-AI/Wan2.1-VACE-14B:diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-VACE-14B:models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-VACE-14B:Wan2.1_VAE.pth" \ - --learning_rate 1e-4 \ + --learning_rate 5e-5 \ --num_epochs 5 \ --remove_prefix_in_ckpt "pipe.vace." \ --output_path "./models/train/Wan2.1-VACE-14B_lora" \ @@ -15,4 +15,5 @@ accelerate launch examples/wanvideo/model_training/train.py \ --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 \ No newline at end of file + --use_gradient_checkpointing_offload +# The learning rate is kept consistent with the settings in the original paper \ No newline at end of file From 70f531b724b6c5588a71334c27304dec7337f7be Mon Sep 17 00:00:00 2001 From: Artiprocher Date: Mon, 19 Jan 2026 17:37:30 +0800 Subject: [PATCH 2/3] update wan-vace training scripts --- .../wanvideo/model_training/lora/Wan2.2-VACE-Fun-A14B.sh | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-) diff --git a/examples/wanvideo/model_training/lora/Wan2.2-VACE-Fun-A14B.sh b/examples/wanvideo/model_training/lora/Wan2.2-VACE-Fun-A14B.sh index 93b38cf..916752b 100644 --- a/examples/wanvideo/model_training/lora/Wan2.2-VACE-Fun-A14B.sh +++ b/examples/wanvideo/model_training/lora/Wan2.2-VACE-Fun-A14B.sh @@ -7,7 +7,7 @@ accelerate launch examples/wanvideo/model_training/train.py \ --num_frames 17 \ --dataset_repeat 100 \ --model_id_with_origin_paths "PAI/Wan2.2-VACE-Fun-A14B:high_noise_model/diffusion_pytorch_model*.safetensors,PAI/Wan2.2-VACE-Fun-A14B:models_t5_umt5-xxl-enc-bf16.pth,PAI/Wan2.2-VACE-Fun-A14B:Wan2.1_VAE.pth" \ - --learning_rate 1e-4 \ + --learning_rate 5e-5 \ --num_epochs 5 \ --remove_prefix_in_ckpt "pipe.vace." \ --output_path "./models/train/Wan2.2-VACE-Fun-A14B_high_noise_lora" \ @@ -19,6 +19,7 @@ accelerate launch examples/wanvideo/model_training/train.py \ --max_timestep_boundary 0.358 \ --min_timestep_boundary 0 # boundary corresponds to timesteps [900, 1000] +# The learning rate is kept consistent with the settings in the original paper accelerate launch examples/wanvideo/model_training/train.py \ --dataset_base_path data/example_video_dataset \ @@ -29,7 +30,7 @@ accelerate launch examples/wanvideo/model_training/train.py \ --num_frames 17 \ --dataset_repeat 100 \ --model_id_with_origin_paths "PAI/Wan2.2-VACE-Fun-A14B:low_noise_model/diffusion_pytorch_model*.safetensors,PAI/Wan2.2-VACE-Fun-A14B:models_t5_umt5-xxl-enc-bf16.pth,PAI/Wan2.2-VACE-Fun-A14B:Wan2.1_VAE.pth" \ - --learning_rate 1e-4 \ + --learning_rate 5e-5 \ --num_epochs 5 \ --remove_prefix_in_ckpt "pipe.vace." \ --output_path "./models/train/Wan2.2-VACE-Fun-A14B_low_noise_lora" \ @@ -40,4 +41,5 @@ accelerate launch examples/wanvideo/model_training/train.py \ --use_gradient_checkpointing_offload \ --max_timestep_boundary 1 \ --min_timestep_boundary 0.358 -# boundary corresponds to timesteps [0, 900] \ No newline at end of file +# boundary corresponds to timesteps [0, 900] +# The learning rate is kept consistent with the settings in the original paper \ No newline at end of file From 8ad2d9884bbaf450fc88dba917162d14bdbbe1ad Mon Sep 17 00:00:00 2001 From: Artiprocher Date: Mon, 19 Jan 2026 17:43:07 +0800 Subject: [PATCH 3/3] update lr in wan-vace training scripts --- .../model_training/full/Wan2.1-VACE-1.3B-Preview.sh | 5 +++-- examples/wanvideo/model_training/full/Wan2.1-VACE-1.3B.sh | 5 +++-- examples/wanvideo/model_training/full/Wan2.1-VACE-14B.sh | 5 +++-- .../wanvideo/model_training/full/Wan2.2-VACE-Fun-A14B.sh | 8 +++++--- .../model_training/lora/Wan2.1-VACE-1.3B-Preview.sh | 5 ++--- examples/wanvideo/model_training/lora/Wan2.1-VACE-1.3B.sh | 5 ++--- examples/wanvideo/model_training/lora/Wan2.1-VACE-14B.sh | 5 ++--- .../wanvideo/model_training/lora/Wan2.2-VACE-Fun-A14B.sh | 8 +++----- 8 files changed, 23 insertions(+), 23 deletions(-) diff --git a/examples/wanvideo/model_training/full/Wan2.1-VACE-1.3B-Preview.sh b/examples/wanvideo/model_training/full/Wan2.1-VACE-1.3B-Preview.sh index b348874..19b6ecb 100644 --- a/examples/wanvideo/model_training/full/Wan2.1-VACE-1.3B-Preview.sh +++ b/examples/wanvideo/model_training/full/Wan2.1-VACE-1.3B-Preview.sh @@ -7,10 +7,11 @@ accelerate launch examples/wanvideo/model_training/train.py \ --num_frames 49 \ --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 \ + --learning_rate 5e-5 \ --num_epochs 2 \ --remove_prefix_in_ckpt "pipe.vace." \ --output_path "./models/train/Wan2.1-VACE-1.3B-Preview_full" \ --trainable_models "vace" \ --extra_inputs "vace_video,vace_reference_image" \ - --use_gradient_checkpointing_offload \ No newline at end of file + --use_gradient_checkpointing_offload +# The learning rate is kept consistent with the settings in the original paper \ No newline at end of file diff --git a/examples/wanvideo/model_training/full/Wan2.1-VACE-1.3B.sh b/examples/wanvideo/model_training/full/Wan2.1-VACE-1.3B.sh index 763252e..f9768c6 100644 --- a/examples/wanvideo/model_training/full/Wan2.1-VACE-1.3B.sh +++ b/examples/wanvideo/model_training/full/Wan2.1-VACE-1.3B.sh @@ -7,10 +7,11 @@ accelerate launch examples/wanvideo/model_training/train.py \ --num_frames 49 \ --dataset_repeat 100 \ --model_id_with_origin_paths "Wan-AI/Wan2.1-VACE-1.3B:diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-VACE-1.3B:models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-VACE-1.3B:Wan2.1_VAE.pth" \ - --learning_rate 1e-4 \ + --learning_rate 5e-5 \ --num_epochs 2 \ --remove_prefix_in_ckpt "pipe.vace." \ --output_path "./models/train/Wan2.1-VACE-1.3B_full" \ --trainable_models "vace" \ --extra_inputs "vace_video,vace_reference_image" \ - --use_gradient_checkpointing_offload \ No newline at end of file + --use_gradient_checkpointing_offload +# The learning rate is kept consistent with the settings in the original paper \ No newline at end of file diff --git a/examples/wanvideo/model_training/full/Wan2.1-VACE-14B.sh b/examples/wanvideo/model_training/full/Wan2.1-VACE-14B.sh index c549263..401a647 100644 --- a/examples/wanvideo/model_training/full/Wan2.1-VACE-14B.sh +++ b/examples/wanvideo/model_training/full/Wan2.1-VACE-14B.sh @@ -7,10 +7,11 @@ accelerate launch --config_file examples/wanvideo/model_training/full/accelerate --num_frames 17 \ --dataset_repeat 100 \ --model_id_with_origin_paths "Wan-AI/Wan2.1-VACE-14B:diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-VACE-14B:models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-VACE-14B:Wan2.1_VAE.pth" \ - --learning_rate 1e-4 \ + --learning_rate 5e-5 \ --num_epochs 2 \ --remove_prefix_in_ckpt "pipe.vace." \ --output_path "./models/train/Wan2.1-VACE-14B_full" \ --trainable_models "vace" \ --extra_inputs "vace_video,vace_reference_image" \ - --use_gradient_checkpointing_offload \ No newline at end of file + --use_gradient_checkpointing_offload +# The learning rate is kept consistent with the settings in the original paper \ No newline at end of file diff --git a/examples/wanvideo/model_training/full/Wan2.2-VACE-Fun-A14B.sh b/examples/wanvideo/model_training/full/Wan2.2-VACE-Fun-A14B.sh index ecfef32..ba3e875 100644 --- a/examples/wanvideo/model_training/full/Wan2.2-VACE-Fun-A14B.sh +++ b/examples/wanvideo/model_training/full/Wan2.2-VACE-Fun-A14B.sh @@ -7,7 +7,7 @@ accelerate launch --config_file examples/wanvideo/model_training/full/accelerate --num_frames 17 \ --dataset_repeat 100 \ --model_id_with_origin_paths "PAI/Wan2.2-VACE-Fun-A14B:high_noise_model/diffusion_pytorch_model*.safetensors,PAI/Wan2.2-VACE-Fun-A14B:models_t5_umt5-xxl-enc-bf16.pth,PAI/Wan2.2-VACE-Fun-A14B:Wan2.1_VAE.pth" \ - --learning_rate 1e-4 \ + --learning_rate 5e-5 \ --num_epochs 2 \ --remove_prefix_in_ckpt "pipe.vace." \ --output_path "./models/train/Wan2.2-VACE-Fun-A14B_high_noise_full" \ @@ -18,6 +18,7 @@ accelerate launch --config_file examples/wanvideo/model_training/full/accelerate --min_timestep_boundary 0 \ --initialize_model_on_cpu # boundary corresponds to timesteps [900, 1000] +# The learning rate is kept consistent with the settings in the original paper accelerate launch --config_file examples/wanvideo/model_training/full/accelerate_config_14B.yaml examples/wanvideo/model_training/train.py \ @@ -29,7 +30,7 @@ accelerate launch --config_file examples/wanvideo/model_training/full/accelerate --num_frames 17 \ --dataset_repeat 100 \ --model_id_with_origin_paths "PAI/Wan2.2-VACE-Fun-A14B:low_noise_model/diffusion_pytorch_model*.safetensors,PAI/Wan2.2-VACE-Fun-A14B:models_t5_umt5-xxl-enc-bf16.pth,PAI/Wan2.2-VACE-Fun-A14B:Wan2.1_VAE.pth" \ - --learning_rate 1e-4 \ + --learning_rate 5e-5 \ --num_epochs 2 \ --remove_prefix_in_ckpt "pipe.vace." \ --output_path "./models/train/Wan2.2-VACE-Fun-A14B_low_noise_full" \ @@ -39,4 +40,5 @@ accelerate launch --config_file examples/wanvideo/model_training/full/accelerate --max_timestep_boundary 1 \ --min_timestep_boundary 0.358 \ --initialize_model_on_cpu -# boundary corresponds to timesteps [0, 900] \ No newline at end of file +# boundary corresponds to timesteps [0, 900] +# The learning rate is kept consistent with the settings in the original paper \ No newline at end of file diff --git a/examples/wanvideo/model_training/lora/Wan2.1-VACE-1.3B-Preview.sh b/examples/wanvideo/model_training/lora/Wan2.1-VACE-1.3B-Preview.sh index 1f25eef..2bcb55b 100644 --- a/examples/wanvideo/model_training/lora/Wan2.1-VACE-1.3B-Preview.sh +++ b/examples/wanvideo/model_training/lora/Wan2.1-VACE-1.3B-Preview.sh @@ -6,7 +6,7 @@ accelerate launch examples/wanvideo/model_training/train.py \ --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 5e-5 \ + --learning_rate 1e-4 \ --num_epochs 5 \ --remove_prefix_in_ckpt "pipe.vace." \ --output_path "./models/train/Wan2.1-VACE-1.3B-Preview_lora" \ @@ -14,5 +14,4 @@ accelerate launch examples/wanvideo/model_training/train.py \ --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 -# The learning rate is kept consistent with the settings in the original paper \ No newline at end of file + --use_gradient_checkpointing_offload \ No newline at end of file diff --git a/examples/wanvideo/model_training/lora/Wan2.1-VACE-1.3B.sh b/examples/wanvideo/model_training/lora/Wan2.1-VACE-1.3B.sh index c8b77cc..b565078 100644 --- a/examples/wanvideo/model_training/lora/Wan2.1-VACE-1.3B.sh +++ b/examples/wanvideo/model_training/lora/Wan2.1-VACE-1.3B.sh @@ -6,7 +6,7 @@ accelerate launch examples/wanvideo/model_training/train.py \ --width 832 \ --dataset_repeat 100 \ --model_id_with_origin_paths "Wan-AI/Wan2.1-VACE-1.3B:diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-VACE-1.3B:models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-VACE-1.3B:Wan2.1_VAE.pth" \ - --learning_rate 5e-5 \ + --learning_rate 1e-4 \ --num_epochs 5 \ --remove_prefix_in_ckpt "pipe.vace." \ --output_path "./models/train/Wan2.1-VACE-1.3B_lora" \ @@ -14,5 +14,4 @@ accelerate launch examples/wanvideo/model_training/train.py \ --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 -# The learning rate is kept consistent with the settings in the original paper \ No newline at end of file + --use_gradient_checkpointing_offload \ No newline at end of file diff --git a/examples/wanvideo/model_training/lora/Wan2.1-VACE-14B.sh b/examples/wanvideo/model_training/lora/Wan2.1-VACE-14B.sh index 28bd05c..633ea0e 100644 --- a/examples/wanvideo/model_training/lora/Wan2.1-VACE-14B.sh +++ b/examples/wanvideo/model_training/lora/Wan2.1-VACE-14B.sh @@ -7,7 +7,7 @@ accelerate launch examples/wanvideo/model_training/train.py \ --num_frames 17 \ --dataset_repeat 100 \ --model_id_with_origin_paths "Wan-AI/Wan2.1-VACE-14B:diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-VACE-14B:models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-VACE-14B:Wan2.1_VAE.pth" \ - --learning_rate 5e-5 \ + --learning_rate 1e-4 \ --num_epochs 5 \ --remove_prefix_in_ckpt "pipe.vace." \ --output_path "./models/train/Wan2.1-VACE-14B_lora" \ @@ -15,5 +15,4 @@ accelerate launch examples/wanvideo/model_training/train.py \ --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 -# The learning rate is kept consistent with the settings in the original paper \ No newline at end of file + --use_gradient_checkpointing_offload \ No newline at end of file diff --git a/examples/wanvideo/model_training/lora/Wan2.2-VACE-Fun-A14B.sh b/examples/wanvideo/model_training/lora/Wan2.2-VACE-Fun-A14B.sh index 916752b..93b38cf 100644 --- a/examples/wanvideo/model_training/lora/Wan2.2-VACE-Fun-A14B.sh +++ b/examples/wanvideo/model_training/lora/Wan2.2-VACE-Fun-A14B.sh @@ -7,7 +7,7 @@ accelerate launch examples/wanvideo/model_training/train.py \ --num_frames 17 \ --dataset_repeat 100 \ --model_id_with_origin_paths "PAI/Wan2.2-VACE-Fun-A14B:high_noise_model/diffusion_pytorch_model*.safetensors,PAI/Wan2.2-VACE-Fun-A14B:models_t5_umt5-xxl-enc-bf16.pth,PAI/Wan2.2-VACE-Fun-A14B:Wan2.1_VAE.pth" \ - --learning_rate 5e-5 \ + --learning_rate 1e-4 \ --num_epochs 5 \ --remove_prefix_in_ckpt "pipe.vace." \ --output_path "./models/train/Wan2.2-VACE-Fun-A14B_high_noise_lora" \ @@ -19,7 +19,6 @@ accelerate launch examples/wanvideo/model_training/train.py \ --max_timestep_boundary 0.358 \ --min_timestep_boundary 0 # boundary corresponds to timesteps [900, 1000] -# The learning rate is kept consistent with the settings in the original paper accelerate launch examples/wanvideo/model_training/train.py \ --dataset_base_path data/example_video_dataset \ @@ -30,7 +29,7 @@ accelerate launch examples/wanvideo/model_training/train.py \ --num_frames 17 \ --dataset_repeat 100 \ --model_id_with_origin_paths "PAI/Wan2.2-VACE-Fun-A14B:low_noise_model/diffusion_pytorch_model*.safetensors,PAI/Wan2.2-VACE-Fun-A14B:models_t5_umt5-xxl-enc-bf16.pth,PAI/Wan2.2-VACE-Fun-A14B:Wan2.1_VAE.pth" \ - --learning_rate 5e-5 \ + --learning_rate 1e-4 \ --num_epochs 5 \ --remove_prefix_in_ckpt "pipe.vace." \ --output_path "./models/train/Wan2.2-VACE-Fun-A14B_low_noise_lora" \ @@ -41,5 +40,4 @@ accelerate launch examples/wanvideo/model_training/train.py \ --use_gradient_checkpointing_offload \ --max_timestep_boundary 1 \ --min_timestep_boundary 0.358 -# boundary corresponds to timesteps [0, 900] -# The learning rate is kept consistent with the settings in the original paper \ No newline at end of file +# boundary corresponds to timesteps [0, 900] \ No newline at end of file