配置要求

我在本地使用了
官方的案例：
DiffSynth-Studio\examples\wanvideo\model_training\lora\Wan2.2-T2V-A14B.sh
数据也是官方的案例

进行了训练 视频lora训练

我的配置是
cpu： intel(R) core(TM) Ultra 9 285K (24核)
gpu： nvidia 5090 32G
内存：96G

但是它没有进度 就像卡住了一样

日志信息：
? 已激活虚拟环境，正在启动训练...

🚀 启动 accelerate 训练（通过 Python 构造命令）
📁 当前工作目录: G:\H\DiffSynth-Studio
🐍 Python: G:\H\DiffSynth-Studio_env\Scripts\python.exe

🔧 构造的命令:
accelerate launch --config_file C:\Users\yi.cache\huggingface\accelerate\default_config.yaml examples/wanvideo/model_training/train.py --dataset_base_path data/example_video_dataset --dataset_metadata_path data/example_video_dataset/metadata.csv --dataset_repeat 100 --dataset_num_workers 0 --model_id_with_origin_paths Wan-AI/Wan2.2-T2V-A14B:high_noise_model/diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.2-T2V-A14B:models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.2-T2V-A14B:Wan2.1_VAE.pth --learning_rate 1e-4 --num_epochs 5 --output_path ./models/train/Wan2.2-T2V-A14B_high_noise_lora --remove_prefix_in_ckpt pipe.dit. --lora_base_model dit --lora_target_modules q,k,v,o,ffn.0,ffn.2 --lora_rank 32 --gradient_accumulation_steps 4 --height 480 --width 832 --num_frames 49 --max_timestep_boundary 0.417 --min_timestep_boundary 0

⏳ 开始执行 accelerate 训练...

G:\H\DiffSynth-Studio\examples\wanvideo\model_training\train.py:30: UserWarning: Gradient checkpointing is detected as disabled. To prevent out-of-memory errors, the training framework will forcibly enable gradient checkpointing.
warnings.warn("Gradient checkpointing is detected as disabled. To prevent out-of-memory errors, the training framework will forcibly enable gradient checkpointing.")
To avoid repeatedly downloading model files, (Wan-AI/Wan2.2-T2V-A14B, models_t5_umt5-xxl-enc-bf16.pth) is redirected to ('DiffSynth-Studio/Wan-Series-Converted-Safetensors', 'models_t5_umt5-xxl-enc-bf16.safetensors'). You can use redirect_common_files=False to disable file redirection.
To avoid repeatedly downloading model files, (Wan-AI/Wan2.2-T2V-A14B, Wan2.1_VAE.pth) is redirected to ('DiffSynth-Studio/Wan-Series-Converted-Safetensors', 'Wan2.1_VAE.safetensors'). You can use redirect_common_files=False to disable file redirection.
Downloading Model from https://www.modelscope.cn to directory: G:\H\DiffSynth-Studio\models\Wan-AI\Wan2.2-T2V-A14B
2026-01-16 23:07:34,727 - modelscope - INFO - Target directory already exists, skipping creation.
Loading models from: [
"./models\Wan-AI/Wan2.2-T2V-A14B\high_noise_model\diffusion_pytorch_model-00001-of-00006.safetensors",
"./models\Wan-AI/Wan2.2-T2V-A14B\high_noise_model\diffusion_pytorch_model-00002-of-00006.safetensors",
"./models\Wan-AI/Wan2.2-T2V-A14B\high_noise_model\diffusion_pytorch_model-00003-of-00006.safetensors",
"./models\Wan-AI/Wan2.2-T2V-A14B\high_noise_model\diffusion_pytorch_model-00004-of-00006.safetensors",
"./models\Wan-AI/Wan2.2-T2V-A14B\high_noise_model\diffusion_pytorch_model-00005-of-00006.safetensors",
"./models\Wan-AI/Wan2.2-T2V-A14B\high_noise_model\diffusion_pytorch_model-00006-of-00006.safetensors"
]
Loaded model: {
"model_name": "wan_video_dit",
"model_class": "diffsynth.models.wan_video_dit.WanModel",
"extra_kwargs": {
"has_image_input": false,
"patch_size": [
1,
2,
2
],
"in_dim": 16,
"dim": 5120,
"ffn_dim": 13824,
"freq_dim": 256,
"text_dim": 4096,
"out_dim": 16,
"num_heads": 40,
"num_layers": 40,
"eps": 1e-06
}
}
Downloading Model from https://www.modelscope.cn to directory: G:\H\DiffSynth-Studio\models\DiffSynth-Studio\Wan-Series-Converted-Safetensors
Loading models from: "./models\DiffSynth-Studio/Wan-Series-Converted-Safetensors\models_t5_umt5-xxl-enc-bf16.safetensors"
Loaded model: {
"model_name": "wan_video_text_encoder",
"model_class": "diffsynth.models.wan_video_text_encoder.WanTextEncoder",
"extra_kwargs": null
}
Downloading Model from https://www.modelscope.cn to directory: G:\H\DiffSynth-Studio\models\DiffSynth-Studio\Wan-Series-Converted-Safetensors
Loading models from: "./models\DiffSynth-Studio/Wan-Series-Converted-Safetensors\Wan2.1_VAE.safetensors"
Loaded model: {
"model_name": "wan_video_vae",
"model_class": "diffsynth.models.wan_video_vae.WanVideoVAE",
"extra_kwargs": null
}
Using wan_video_text_encoder from "./models\DiffSynth-Studio/Wan-Series-Converted-Safetensors\models_t5_umt5-xxl-enc-bf16.safetensors".
Using wan_video_dit from [
"./models\Wan-AI/Wan2.2-T2V-A14B\high_noise_model\diffusion_pytorch_model-00001-of-00006.safetensors",
"./models\Wan-AI/Wan2.2-T2V-A14B\high_noise_model\diffusion_pytorch_model-00002-of-00006.safetensors",
"./models\Wan-AI/Wan2.2-T2V-A14B\high_noise_model\diffusion_pytorch_model-00003-of-00006.safetensors",
"./models\Wan-AI/Wan2.2-T2V-A14B\high_noise_model\diffusion_pytorch_model-00004-of-00006.safetensors",
"./models\Wan-AI/Wan2.2-T2V-A14B\high_noise_model\diffusion_pytorch_model-00005-of-00006.safetensors",
"./models\Wan-AI/Wan2.2-T2V-A14B\high_noise_model\diffusion_pytorch_model-00006-of-00006.safetensors"
].
Using wan_video_vae from "./models\DiffSynth-Studio/Wan-Series-Converted-Safetensors\Wan2.1_VAE.safetensors".
No wan_video_image_encoder models available. This is not an error.
No wan_video_motion_controller models available. This is not an error.
No wan_video_vace models available. This is not an error.
No wan_video_vap models available. This is not an error.
No wans2v_audio_encoder models available. This is not an error.
No wan_video_animate_adapter models available. This is not an error.
Downloading Model from https://www.modelscope.cn to directory: G:\H\DiffSynth-Studio\models\Wan-AI\Wan2.1-T2V-1.3B
2026-01-16 23:08:11,204 - modelscope - INFO - Target directory already exists, skipping creation.
Downloading Model from https://www.modelscope.cn to directory: G:\H\DiffSynth-Studio\models\Wan-AI\Wan2.2-S2V-14B
2026-01-16 23:08:13,681 - modelscope - INFO - Target directory already exists, skipping creation.
0%| | 0/100 [00:00<?, ?it/s]

30分钟没有变化

Start with examples/wanvideo/model_training/lora/Wan2.2-T2V-A14B.sh and examples/wanvideo/model_training/train.py, then reproduce the reported command with the example dataset and settings. Trace the transition after model loading and the 0/100 progress display; done means the stall is explained and training either proceeds or reports a clear actionable failure.