为什么加载不到权重路径

[JerryZ01]

具体报错信息
Traceback (most recent call last):
File "/root/LongVie/inference.py", line 92, in
main(args)
File "/root/LongVie/inference.py", line 24, in main
pipe = LongViePipeline.from_pretrained(
File "/root/LongVie/diffsynth/pipelines/wan_video_new_longvie.py", line 407, in from_pretrained
model_manager.load_model(
File "/root/LongVie/diffsynth/models/model_manager.py", line 408, in load_model
if model_detector.match(file_path, state_dict):
File "/root/LongVie/diffsynth/models/model_manager.py", line 166, in match
state_dict = load_state_dict(file_path)
File "/root/LongVie/diffsynth/models/utils.py", line 66, in load_state_dict
if file_path.endswith(".safetensors"):
AttributeError: 'list' object has no attribute 'endswith'

[JianxGao]

你好，在你的截图里我没有找到wan2.1-14b-i2v-480p的weights，可能需要check一下这个

[JerryZ01]

你好，在你的截图里我没有找到wan2.1-14b-i2v-480p的weights，可能需要check一下这个

什么意思，我看redeme就执行一下脚本就好了，没说在哪配置wan2.1的权重

[JianxGao]

运行：

python download_wan2.1.py

参考readme这里：
https://github.com/Vchitect/LongVie#download-weights

[JerryZ01]

运行：

python download_wan2.1.py
参考readme在这里： https://github.com/Vchitect/LongVie#download-weights

下载了wan2.1执行还是以下的问题
directly, please report this to the maintainers of the package that installed pynvml for you.
import pynvml # type: ignore[import]
Loading models from: []
Traceback (most recent call last):
File "/root/LongVie/inference.py", line 92, in
main(args)
File "/root/LongVie/inference.py", line 24, in main
pipe = LongViePipeline.from_pretrained(
File "/root/LongVie/diffsynth/pipelines/wan_video_new_longvie.py", line 407, in from_pretrained
model_manager.load_model(
File "/root/LongVie/diffsynth/models/model_manager.py", line 408, in load_model
if model_detector.match(file_path, state_dict):
File "/root/LongVie/diffsynth/models/model_manager.py", line 166, in match
state_dict = load_state_dict(file_path)
File "/root/LongVie/diffsynth/models/utils.py", line 66, in load_state_dict
if file_path.endswith(".safetensors"):
AttributeError: 'list' object has no attribute 'endswith'

[JianxGao]

你可以试试下面的方案，来自这个issue #5 (comment)

# In diffsynth/models/utils.py

def load_state_dict(file_path, torch_dtype=None, device="cpu"):
    # --- ADDED THIS BLOCK ---
    if isinstance(file_path, list):
        state_dict = {}
        for f in file_path:
            state_dict.update(load_state_dict(f, torch_dtype=torch_dtype, device=device))
        return state_dict
    # ------------------------
    
    if file_path.endswith(".safetensors"):
        return load_state_dict_from_safetensors(file_path, torch_dtype=torch_dtype, device=device)
    else:
        return load_state_dict_from_bin(file_path, torch_dtype=torch_dtype, device=device)

[JerryZ01]

你可以尝试下面的方案，来自这个issue #5 (评论)

In diffsynth/models/utils.py

def load_state_dict(file_path, torch_dtype=None, device="cpu"):
# --- ADDED THIS BLOCK ---
if isinstance(file_path, list):
state_dict = {}
for f in file_path:
state_dict.update(load_state_dict(f, torch_dtype=torch_dtype, device=device))
return state_dict
# ------------------------

if file_path.endswith(".safetensors"):
    return load_state_dict_from_safetensors(file_path, torch_dtype=torch_dtype, device=device)
else:
    return load_state_dict_from_bin(file_path, torch_dtype=torch_dtype, device=device)

感谢，我按照你的方式和附带的issue解决了list的问题，重新执行日志如下，好像还是没成功
(longvie) root@di-20251224142855-vvbfl:~/LongVie# bash sample_longvideo.sh /root/miniconda3/envs/longvie/lib/python3.10/site-packages/torch/cuda/__init__.py:61: FutureWarning: The pynvml package is deprecated. Please install nvidia-ml-py instead. If you did not install pynvml directly, please report this to the maintainers of the package that installed pynvml for you. import pynvml # type: ignore[import] Loading models from: ['./models/Wan-AI/Wan2.1-I2V-14B-480P/diffusion_pytorch_model-00006-of-00007.safetensors', './models/Wan-AI/Wan2.1-I2V-14B-480P/diffusion_pytorch_model-00005-of-00007.safetensors', './models/Wan-AI/Wan2.1-I2V-14B-480P/diffusion_pytorch_model-00004-of-00007.safetensors', './models/Wan-AI/Wan2.1-I2V-14B-480P/diffusion_pytorch_model-00002-of-00007.safetensors', './models/Wan-AI/Wan2.1-I2V-14B-480P/diffusion_pytorch_model-00001-of-00007.safetensors', './models/Wan-AI/Wan2.1-I2V-14B-480P/diffusion_pytorch_model-00003-of-00007.safetensors', './models/Wan-AI/Wan2.1-I2V-14B-480P/diffusion_pytorch_model-00007-of-00007.safetensors'] model_name: wan_video_dit model_class: WanModel This model is initialized with extra kwargs: {'has_image_input': True, 'patch_size': [1, 2, 2], 'in_dim': 36, 'dim': 5120, 'ffn_dim': 13824, 'freq_dim': 256, 'text_dim': 4096, 'out_dim': 16, 'num_heads': 40, 'num_layers': 40, 'eps': 1e-06} The following models are loaded: ['wan_video_dit']. Loading models from: ./models/Wan-AI/Wan2.1-I2V-14B-480P/models_t5_umt5-xxl-enc-bf16.pth model_name: wan_video_text_encoder model_class: WanTextEncoder The following models are loaded: ['wan_video_text_encoder']. Loading models from: ./models/Wan-AI/Wan2.1-I2V-14B-480P/Wan2.1_VAE.pth model_name: wan_video_vae model_class: WanVideoVAE The following models are loaded: ['wan_video_vae']. Loading models from: ./models/Wan-AI/Wan2.1-I2V-14B-480P/models_clip_open-clip-xlm-roberta-large-vit-huge-14.pth model_name: wan_video_image_encoder model_class: WanImageEncoder The following models are loaded: ['wan_video_image_encoder']. Using wan_video_text_encoder from ./models/Wan-AI/Wan2.1-I2V-14B-480P/models_t5_umt5-xxl-enc-bf16.pth. Using wan_video_dit from ['./models/Wan-AI/Wan2.1-I2V-14B-480P/diffusion_pytorch_model-00006-of-00007.safetensors', './models/Wan-AI/Wan2.1-I2V-14B-480P/diffusion_pytorch_model-00005-of-00007.safetensors', './models/Wan-AI/Wan2.1-I2V-14B-480P/diffusion_pytorch_model-00004-of-00007.safetensors', './models/Wan-AI/Wan2.1-I2V-14B-480P/diffusion_pytorch_model-00002-of-00007.safetensors', './models/Wan-AI/Wan2.1-I2V-14B-480P/diffusion_pytorch_model-00001-of-00007.safetensors', './models/Wan-AI/Wan2.1-I2V-14B-480P/diffusion_pytorch_model-00003-of-00007.safetensors', './models/Wan-AI/Wan2.1-I2V-14B-480P/diffusion_pytorch_model-00007-of-00007.safetensors']. Using wan_video_vae from ./models/Wan-AI/Wan2.1-I2V-14B-480P/Wan2.1_VAE.pth. Using wan_video_image_encoder from ./models/Wan-AI/Wan2.1-I2V-14B-480P/models_clip_open-clip-xlm-roberta-large-vit-huge-14.pth. No wan_video_motion_controller models available. No wan_video_vace models available. No wans2v_audio_encoder models available. No wan_video_animate_adapter models available. Load Dual Controller! Updata self-attention weight in DiT! Downloading Model from https://www.modelscope.cn to directory: /root/LongVie/models/Wan-AI/Wan2.1-T2V-1.3B 2026-01-07 10:07:47,563 - modelscope - INFO - Target directory already exists, skipping creation. 0%| | 0/50 [00:00<?, ?it/s]sample_longvideo.sh: line 7: 3435587 Floating point exceptionpython inference.py --json_file /root/LongVie/example/ride_horse/cond.json --image_path /root/LongVie/example/ride_horse/first.png --video_name ride_horse --control_weight_path /root/LongVie/models/LongVie/control.safetensors --dit_weight_path /root/LongVie/models/LongVie/dit.safetensors

我的机器是H20，貌似报Floating point exceptionpython

[JianxGao]

这个可能是H20的错误，搜了一下相似的，感觉你可以参考一下
aigc-apps/EasyAnimate#32
PaddlePaddle/Paddle#66921

[yyang181]

H20不支持bf16, 直接强行更新Cublas到最新版本就可以了：pip install nvidia-cublas-cu12 -U

[JerryZ01]

这个可能是H20的错误，搜相似的，感觉你可以参考一下 aigc-apps/EasyAnimate#32 PaddlePaddle/Paddle#66921

感谢大佬解答，这个推理支持分布式推理嘛，跑了官方的样例，感觉好慢

[JianxGao]

暂时还不支持多卡推理，我会尽快处理这个

[JerryZ01]

暂时不支持多卡推理，我会加速处理这个

我可以用自己的视频去生成嘛，需要向你们代码example里一样构建出depth和track然后去推理嘛

[JianxGao]

是的，需要准备好depth track和prompt

[JerryZ01]

是的，需要准备好深度跟踪和提示
执行get_depth.py报错了，video_depth_anything_vitl.pth这个东西在哪，我看代码里没有呢
/root/LongVie/utils/get_depth.py:76: FutureWarning: You are using torch.loadwithweights_only=False(the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value forweights_onlywill be flipped toTrue. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via torch.serialization.add_safe_globals. We recommend you start setting weights_only=True for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature. video_depth_anything.load_state_dict(torch.load(f'./checkpoints/video_depth_anything_{args.encoder}.pth', map_location='cpu'), strict=True) Traceback (most recent call last): File "/root/LongVie/utils/get_depth.py", line 76, in <module> video_depth_anything.load_state_dict(torch.load(f'./checkpoints/video_depth_anything_{args.encoder}.pth', map_location='cpu'), strict=True) File "/root/miniconda3/envs/longvie/lib/python3.10/site-packages/torch/serialization.py", line 1319, in load with _open_file_like(f, "rb") as opened_file: File "/root/miniconda3/envs/longvie/lib/python3.10/site-packages/torch/serialization.py", line 659, in _open_file_like return _open_file(name_or_buffer, mode) File "/root/miniconda3/envs/longvie/lib/python3.10/site-packages/torch/serialization.py", line 640, in __init__ super().__init__(open(name, mode)) FileNotFoundError: [Errno 2] No such file or directory: './checkpoints/video_depth_anything_vitl.pth'

[JianxGao]

提取depth和track都需要下载对应模型的ckpt：
depth: https://github.com/DepthAnything/Video-Depth-Anything#pre-trained-models, (Video-Depth-Anything-Large)
track: https://github.com/henry123-boy/SpaTracker#data

[JianxGao]

Hi, @JerryZ01
LongVie2现在支持多卡推理了。因我们在controlnet的attention head数量是20，所以目前实测下来只能支持2, 4, 5卡这三种并行方式。

你可以用如下指令：

torchrun --master_port=22519 --nproc_per_node=4 inference.py \
    --json_file  ./example/ride_horse/cond.json \
    --image_path ./example/ride_horse/first.png  \
    --video_name ride_horse \
    --control_weight_path ./models/LongVie/control.safetensors \
    --dit_weight_path ./models/LongVie/dit.safetensors \
    --use_usp