【pytorch】加载模型出现问题 Unexpected key(s) in state_dict: “params“, “params_ema“

最新推荐文章于 2023-12-12 15:48:18 发布
原创 最新推荐文章于 2023-12-12 15:48:18 发布 · 1.6k 阅读 · 1 ·
CC 4.0 BY-SA版权
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
文章标签：

#pytorch #深度学习 #人工智能

在尝试加载模型时遇到错误，状态字典缺少如conv_first.weight等键，同时存在未预期的键params和params_ema。文章指出，用户只需关注params_ema内的参数文件，通过修改加载语句，只加载这部分参数到模型中，可以解决报错。
部署运行你感兴趣的模型镜像

最开始

model.load_state_dict(torch.load(model_path), strict=True)

加载模型报错如下

Missing key(s) in state_dict: "conv_first.weight", "conv_first.bias", .... 
Unexpected key(s) in state_dict: "params", "params_ema"

训练好的模型键值有许多，只需要模型参数文件，我的参数文件在params_ema内
修改如下即可

state_dict = torch.load(model_path)
model.load_state_dict(state_dict['params_ema'], strict=True)

您可能感兴趣的与本文相关的镜像

PyTorch 2.9

PyTorch 2.9

PyTorch
Cuda

PyTorch 是一个开源的 Python 机器学习库，基于 Torch 库，底层由 C++ 实现，应用于人工智能领域，如计算机视觉和自然语言处理

【图像超分】论文复现：纯CNN模型！多级大核卷积注意力！MAN的Pytorch源码复现，跑通源码，获得与论文一致的指标和超分可视化结果，各组件可作为即插即用模块积累！
主要更新底层视觉（去噪、超分等）相关的科研内容，形式为【论文精读】+【论文复现】
05-09 645
本文介绍了如何实现和测试多尺度注意网络（MAN）用于单幅图像超分辨率。文章详细说明了从环境配置、数据集准备到模型测试和训练的完整流程。首先，读者需要下载并配置BasicSR项目，并准备好训练和测试数据集。接着，文章指导如何将MAN源码整合到BasicSR中，并提供了不同版本（tiny、light、base）的配置文件。通过执行相应的测试命令，可以获得与论文一致的超分结果和指标。此外，文章还提供了训练配置文件的设置方法，帮助读者进行模型训练。完整代码和预训练模型可通过订阅专栏免费获取。
GFPGAN源代码分析（六）
qq_45969525的博客
12-27 1422
2021SC@SDUSC 一、分析的代码片段 1.代码展示 if different_w: linear_out_channel = (int(math.log(out_size, 2)) * 2 - 2) * num_style_feat else: linear_out_channel = num_style_feat self.final_linear = nn.Linear(channels['4']
GFPGAN源码分析—第三篇
Vaifer233的博客
12-06 2754
源码：utils.py 本篇主要分析utils.py中的class GFOGANer ( ) 1.获取当前项目路径 ROOT_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__))) os.path.abspath(file)获取当前文件的绝对路径C:\Users\Vaifer\Desktop\GFPGAN-v.0.2.1\gfpgan\utils.py os.path.dirname()再获取该文件所在的目录路径C:\
GFPGAN源代码分析（九）
qq_45969525的博客
12-27 943
2021SC@SDUSC 一、分析的代码片段 1.代码展示 class GFPGANer(): def __init__(self, model_path, upscale=2, arch='clean', channel_multiplier=2, bg_upsampler=None): self.upscale = upscale self.bg_upsampler = bg_upsampler # initialize model
解决torch.load中Unexpected key(s) in state_dict的报错问题
weixin_51007334的博客
08-22 2129
今天在使用强化学习跑实验的时候，使用到了这个函数，从已经保存的pth文件中加载模型权重。但是加载的过程中，出现了错误。
Unexpected key(s) in state_dict: “model“
weiyuangong的博客
05-26 1622
问题描述：加载预训练权重时报错Unexpected key(s) in state_dict: "model" weight_dic_path = 'convnext_base_1k_224.pth' state_dic=torch.load(weight_dic_path,map_location=device) model.load_state_dict(state_dic) 解决方法： weight_dic_path = 'convnext_base_1k_224.pth' state_dic
“torch.load“中出现的“Unexpected key(s) in state_dict“报错问题
weixin_43723625的博客
10-17 5246
允许加载过程中出现不匹配的键。但请注意,仍然需要确保模型中的主要参数能够正确加载，以确保模型的有效性。
pytorch使用torch.load加载模型出现错误Unexpected key(s) in state_dict
风影的博客
09-04 3646
记录错误积累经验 猛然来这么一大串错误信息让我有点震惊。 pytorch模型文件以key，value形式保存，现在模型所有key都对不上，有点奇怪。细看发现加载的模型相较于当前模型其key值前面多了“module."。 初步诊断 class Tester(object): TestParams = TestParams def __init__(self, model, test_params): assert isinstance(test_params, TestP
RuntimeError: Error(s) in loading state_dict for ArSSR: Unexpected key(s) in state_dict:
12-29
### 解决 PyTorch 加载 ArSSR 模型时遇到的 `Unexpected key(s) in state_dict` 错误 当加载预训练模型权重时，如果出现 `Unexpected key(s) in state_dict` 的错误提示，通常意味着保存的模型状态字典中的键名与...
RuntimeError: Error(s) in loading state_dict for VMUNet: Missing key(s) in state_dict: "total_ops", "total_params", "vmunet.total_ops", "vmunet.total_params", "vmunet.patch_embed.total_ops", "vmunet.patch_embed.total_params", "vmunet.patch_embed.proj.weight", "vmunet.patch_embed.proj.bias", "vmunet.patch_embed.norm.weight", "vmunet.patch_embed.norm.bias", "vmunet.layers.total_ops", "vmunet.layers.total_params", "vmunet.layers.0.total_ops", "vmunet.layers.0.total_params", "vmunet.layers.0.blocks.total_ops", "vmunet.layers.0.blocks.total_params", "vmunet.layers.0.blocks.0.total_ops", "vmunet.layers.0.blocks.0.total_params", "vmunet.layers.0.blocks.0.ln_1.weight", "vmunet.layers.0.blocks.0.ln_1.bias", "vmunet.layers.0.blocks.0.self_attention.x_proj_weight", "vmunet.layers.0.blocks.0.self_attention.dt_projs_weight", "vmunet.layers.0.blocks.0.self_attention.dt_projs_bias", "vmunet.layers.0.blocks.0.self_attention.A_logs", "vmunet.layers.0.blocks.0.self_attention.Ds", "vmunet.layers.0.blocks.0.self_attention.total_ops", "vmunet.layers.0.blocks.0.self_attention.total_params", "vmunet.layers.0.blocks.0.self_attention.in_proj.weight", "vmunet.layers.0.blocks.0.self_attention.conv2d.weight", "vmunet.layers.0.blocks.0.self_attention.conv2d.bias", "vmunet.layers.0.blocks.0.self_attention.act.total_ops", "vmunet.layers.0.blocks.0.self_attention.act.total_params", "vmunet.layers.0.blocks.0.self_attention.out_norm.weight", "vmunet.layers.0.blocks.0.self_attention.out_norm.bias", "vmunet.layers.0.blocks.0.self_attention.out_proj.weight", "vmunet.layers.0.blocks.0.drop_path.total_ops", "vmunet.layers.0.blocks.0.drop_path.total_params", "vmunet.layers.0.blocks.1.total_ops", "vmunet.layers.0.blocks.1.total_params", "vmunet.layers.0.blocks.1.ln_1.weight", "vmunet.layers.0.blocks.1.ln_1.bias", "vmunet.layers.0.blocks.1.self_attention.x_proj_weight", "vmunet.layers.0.blocks.1.self_attention.dt_projs_weight", "vmunet.layers.0.blocks.1.self_attention.dt_projs_bias", "vmunet.layers.0.blocks.1.self_attention.A_logs", "vmunet.layers.0.blocks.1.self_attention.Ds", "vmunet.layers.0.blocks.1.self_attention.total_ops", "vmunet.layers.0.blocks.1.self_attention.total_params", "vmunet.layers.0.blocks.1.self_attention.in_proj.weight", "vmunet.layers.0.blocks.1.self_attention.conv2d.weight", "vmunet.layers.0.blocks.1.self_attention.conv2d.bias", "vmunet.layers.0.blocks.1.self_attention.act.total_ops", "vmunet.layers.0.blocks.1.self_attention.act.total_params", "vmunet.layers.0.blocks.1.self_attention.out_norm.weight", "vmunet.layers.0.blocks.1.self_attention.out_norm.bias", "vmunet.layers.0.blocks.1.self_attention.out_proj.weight", "vmunet.layers.0.blocks.1.drop_path.total_ops", "vmunet.layers.0.blocks.1.drop_path.total_params", "vmunet.layers.0.downsample.total_ops", "vmunet.layers.0.downsample.total_params", "vmunet.layers.0.downsample.reduction.weight", "vmunet.layers.0.downsample.norm.weight", "vmunet.layers.0.downsample.norm.bias", "vmunet.layers.1.total_ops", "vmunet.layers.1.total_params", "vmunet.layers.1.blocks.total_ops", "vmunet.layers.1.blocks.total_params", "vmunet.layers.1.blocks.0.total_ops", "vmunet.layers.1.blocks.0.total_params", "vmunet.layers.1.blocks.0.ln_1.weight", "vmunet.layers.1.blocks.0.ln_1.bias", "vmunet.layers.1.blocks.0.self_attention.x_proj_weight", "vmunet.layers.1.blocks.0.self_attention.dt_projs_weight", "vmunet.layers.1.blocks.0.self_attention.dt_projs_bias", "vmunet.layers.1.blocks.0.self_attention.A_logs", "vmunet.layers.1.blocks.0.self_attention.Ds", "vmunet.layers.1.blocks.0.self_attention.total_ops", "vmunet.layers.1.blocks.0.self_attention.total_params", "vmunet.layers.1.blocks.0.self_attention.in_proj.weight", "vmunet.layers.1.blocks.0.self_attention.conv2d.weight", "vmunet.layers.1.blocks.0.self_attention.conv2d.bias", "vmunet.layers.1.blocks.0.self_attention.act.total_ops", "vmunet.layers.1.blocks.0.self_attention.act.total_params", "vmunet.layers.1.blocks.0.self_attention.out_norm.weight", "vmunet.layers.1.blocks.0.self_attention.out_norm.bias", "vmunet.layers.1.blocks.0.self_attention.out_proj.weight", "vmunet.layers.1.blocks.0.drop_path.total_ops", "vmunet.layers.1.blocks.0.drop_path.total_params", "vmunet.layers.1.blocks.1.total_ops", "vmunet.layers.1.blocks.1.total_params", "vmunet.layers.1.blocks.1.ln_1.weight", "vmunet.layers.1.blocks.1.ln_1.bias", "vmunet.layers.1.blocks.1.self_attention.x_proj_weight", "vmunet.layers.1.blocks.1.self_attention.dt_projs_weight", "vmunet.layers.1.blocks.1.self_attention.dt_projs_bias", "vmunet.layers.1.blocks.1.self_attention.A_logs", "vmunet.layers.1.blocks.1.self_attention.Ds", "vmunet.layers.1.blocks.1.self_attention.total_ops", "vmunet.layers.1.blocks.1.self_attention.total_params", "vmunet.layers.1.blocks.1.self_attention.in_proj.weight", "vmunet.layers.1.blocks.1.self_attention.conv2d.weight", "vmunet.layers.1.blocks.1.self_attention.conv2d.bias", "vmunet.layers.1.blocks.1.self_attention.act.total_ops", "vmunet.layers.1.blocks.1.self_attention.act.total_params", "vmunet.layers.1.blocks.1.self_attention.out_norm.weight", "vmunet.layers.1.blocks.1.self_attention.out_norm.bias", "vmunet.layers.1.blocks.1.self_attention.out_proj.weight", "vmunet.layers.1.blocks.1.drop_path.total_ops", "vmunet.layers.1.blocks.1.drop_path.total_params", "vmunet.layers.1.downsample.total_ops", "vmunet.layers.1.downsample.total_params", "vmunet.layers.1.downsample.reduction.weight", "vmunet.layers.1.downsample.norm.weight", "vmunet.layers.1.downsample.norm.bias", "vmunet.layers.2.total_ops", "vmunet.layers.2.total_params", "vmunet.layers.2.blocks.total_ops", "vmunet.layers.2.blocks.total_params", "vmunet.layers.2.blocks.0.total_ops", "vmunet.layers.2.blocks.0.total_params", "vmunet.layers.2.blocks.0.ln_1.weight", "vmunet.layers.2.blocks.0.ln_1.bias", "vmunet.layers.2.blocks.0.self_attention.x_proj_weight", "vmunet.layers.2.blocks.0.self_attention.dt_projs_weight", "vmunet.layers.2.blocks.0.self_attention.dt_projs_bias", "vmunet.layers.2.blocks.0.self_attention.A_logs", "vmunet.layers.2.blocks.0.self_attention.Ds", "vmunet.layers.2.blocks.0.self_attention.total_ops", "vmunet.layers.2.blocks.0.self_attention.total_params", "vmunet.layers.2.blocks.0.self_attention.in_proj.weight", "vmunet.layers.2.blocks.0.self_attention.conv2d.weight", "vmunet.layers.2.blocks.0.self_attention.conv2d.bias", "vmunet.layers.2.blocks.0.self_attention.act.total_ops", "vmunet.layers.2.blocks.0.self_attention.act.total_params", "vmunet.layers.2.blocks.0.self_attention.out_norm.weight", "vmunet.layers.2.blocks.0.self_attention.out_norm.bias", "vmunet.layers.2.blocks.0.self_attention.out_proj.weight", "vmunet.layers.2.blocks.0.drop_path.total_ops", "vmunet.layers.2.blocks.0.drop_path.total_params", "vmunet.layers.2.blocks.1.total_ops", "vmunet.layers.2.blocks.1.total_params", "vmunet.layers.2.blocks.1.ln_1.weight", "vmunet.layers.2.blocks.1.ln_1.bias", "vmunet.layers.2.blocks.1.self_attention.x_proj_weight", "vmunet.layers.2.blocks.1.self_attention.dt_projs_weight", "vmunet.layers.2.blocks.1.self_attention.dt_projs_bias", "vmunet.layers.2.blocks.1.self_attention.A_logs", "vmunet.layers.2.blocks.1.self_attention.Ds", "vmunet.layers.2.blocks.1.self_attention.total_ops", "vmunet.layers.2.blocks.1.self_attention.total_params", "vmunet.layers.2.blocks.1.self_attention.in_proj.weight", "vmunet.layers.2.blocks.1.self_attention.conv2d.weight", "vmunet.layers.2.blocks.1.self_attention.conv2d.bias", "vmunet.layers.2.blocks.1.self_attention.act.total_ops", "vmunet.layers.2.blocks.1.self_attention.act.total_params", "vmunet.layers.2.blocks.1.self_attention.out_norm.weight", "vmunet.layers.2.blocks.1.self_attention.out_norm.bias", "vmunet.layers.2.blocks.1.self_attention.out_proj.weight", "vmunet.layers.2.blocks.1.drop_path.total_ops", "vmunet.layers.2.blocks.1.drop_path.total_params", "vmunet.layers.2.downsample.total_ops", "vmunet.layers.2.downsample.total_params", "vmunet.layers.2.downsample.reduction.weight", "vmunet.layers.2.downsample.norm.weight", "vmunet.layers.2.downsample.norm.bias", "vmunet.layers.3.total_ops", "vmunet.layers.3.total_params", "vmunet.layers.3.blocks.total_ops", "vmunet.layers.3.blocks.total_params", "vmunet.layers.3.blocks.0.total_ops", "vmunet.layers.3.blocks.0.total_params", "vmunet.layers.3.blocks.0.ln_1.weight", "vmunet.layers.3.blocks.0.ln_1.bias", "vmunet.layers.3.blocks.0.self_attention.x_proj_weight", "vmunet.layers.3.blocks.0.self_attention.dt_projs_weight", "vmunet.layers.3.blocks.0.self_attention.dt_projs_bias", "vmunet.layers.3.blocks.0.self_attention.A_logs", "vmunet.layers.3.blocks.0.self_attention.Ds", "vmunet.layers.3.blocks.0.self_attention.total_ops", "vmunet.layers.3.blocks.0.self_attention.total_params", "vmunet.layers.3.blocks.0.self_attention.in_proj.weight", "vmunet.layers.3.blocks.0.self_attention.conv2d.weight", "vmunet.layers.3.blocks.0.self_attention.conv2d.bias", "vmunet.layers.3.blocks.0.self_attention.act.total_ops", "vmunet.layers.3.blocks.0.self_attention.act.total_params", "vmunet.layers.3.blocks.0.self_attention.out_norm.weight", "vmunet.layers.3.blocks.0.self_attention.out_norm.bias", "vmunet.layers.3.blocks.0.self_attention.out_proj.weight", "vmunet.layers.3.blocks.0.drop_path.total_ops", "vmunet.layers.3.blocks.0.drop_path.total_params", "vmunet.layers.3.blocks.1.total_ops", "vmunet.layers.3.blocks.1.total_params", "vmunet.layers.3.blocks.1.ln_1.weight", "vmunet.layers.3.blocks.1.ln_1.bias", "vmunet.layers.3.blocks.1.self_attention.x_proj_weight", "vmunet.layers.3.blocks.1.self_attention.dt_projs_weight", "vmunet.layers.3.blocks.1.self_attention.dt_projs_bias", "vmunet.layers.3.blocks.1.self_attention.A_logs", "vmunet.layers.3.blocks.1.self_attention.Ds", "vmunet.layers.3.blocks.1.self_attention.total_ops", "vmunet.layers.3.blocks.1.self_attention.total_params", "vmunet.layers.3.blocks.1.self_attention.in_proj.weight", "vmunet.layers.3.blocks.1.self_attention.conv2d.weight", "vmunet.layers.3.blocks.1.self_attention.conv2d.bias", "vmunet.layers.3.blocks.1.self_attention.act.total_ops", "vmunet.layers.3.blocks.1.self_attention.act.total_params", "vmunet.layers.3.blocks.1.self_attention.out_norm.weight", "vmunet.layers.3.blocks.1.self_attention.out_norm.bias", "vmunet.layers.3.blocks.1.self_attention.out_proj.weight", "vmunet.layers.3.blocks.1.drop_path.total_ops", "vmunet.layers.3.blocks.1.drop_path.total_params", "vmunet.layers_up.total_ops", "vmunet.layers_up.total_params", "vmunet.layers_up.0.total_ops", "vmunet.layers_up.0.total_params", "vmunet.layers_up.0.blocks.total_ops", "vmunet.layers_up.0.blocks.total_params", "vmunet.layers_up.0.blocks.0.total_ops", "vmunet.layers_up.0.blocks.0.total_params", "vmunet.layers_up.0.blocks.0.ln_1.weight", "vmunet.layers_up.0.blocks.0.ln_1.bias", "vmunet.layers_up.0.blocks.0.self_attention.x_proj_weight", "vmunet.layers_up.0.blocks.0.self_attention.dt_projs_weight", "vmunet.layers_up.0.blocks.0.self_attention.dt_projs_bias", "vmunet.layers_up.0.blocks.0.self_attention.A_logs", "vmunet.layers_up.0.blocks.0.self_attention.Ds", "vmunet.layers_up.0.blocks.0.self_attention.total_ops", "vmunet.layers_up.0.blocks.0.self_attention.total_params", "vmunet.layers_up.0.blocks.0.self_attention.in_proj.weight", "vmunet.layers_up.0.blocks.0.self_attention.conv2d.weight", "vmunet.layers_up.0.blocks.0.self_attention.conv2d.bias", "vmunet.layers_up.0.blocks.0.self_attention.act.total_ops", "vmunet.layers_up.0.blocks.0.self_attention.act.total_params", "vmunet.layers_up.0.blocks.0.self_attention.out_norm.weight", "vmunet.layers_up.0.blocks.0.self_attention.out_norm.bias", "vmunet.layers_up.0.blocks.0.self_attention.out_proj.weight", "vmunet.layers_up.0.blocks.0.drop_path.total_ops", "vmunet.layers_up.0.blocks.0.drop_path.total_params", "vmunet.layers_up.0.blocks.1.total_ops", "vmunet.layers_up.0.blocks.1.total_params", "vmunet.layers_up.0.blocks.1.ln_1.weight", "vmunet.layers_up.0.blocks.1.ln_1.bias", "vmunet.layers_up.0.blocks.1.self_attention.x_proj_weight", "vmunet.layers_up.0.blocks.1.self_attention.dt_projs_weight", "vmunet.layers_up.0.blocks.1.self_attention.dt_projs_bias", "vmunet.layers_up.0.blocks.1.self_attention.A_logs", "vmunet.layers_up.0.blocks.1.self_attention.Ds", "vmunet.layers_up.0.blocks.1.self_attention.total_ops", "vmunet.layers_up.0.blocks.1.self_attention.total_params", "vmunet.layers_up.0.blocks.1.self_attention.in_proj.weight", "vmunet.layers_up.0.blocks.1.self_attention.conv2d.weight", "vmunet.layers_up.0.blocks.1.self_attention.conv2d.bias", "vmunet.layers_up.0.blocks.1.self_attention.act.total_ops", "vmunet.layers_up.0.blocks.1.self_attention.act.total_params", "vmunet.layers_up.0.blocks.1.self_attention.out_norm.weight", "vmunet.layers_up.0.blocks.1.self_attention.out_norm.bias", "vmunet.layers_up.0.blocks.1.self_attention.out_proj.weight", "vmunet.layers_up.0.blocks.1.drop_path.total_ops", "vmunet.layers_up.0.blocks.1.drop_path.total_params", "vmunet.layers_up.1.total_ops", "vmunet.layers_up.1.total_params", "vmunet.layers_up.1.blocks.total_ops", "vmunet.layers_up.1.blocks.total_params", "vmunet.layers_up.1.blocks.0.total_ops", "vmunet.layers_up.1.blocks.0.total_params", "vmunet.layers_up.1.blocks.0.ln_1.weight", "vmunet.layers_up.1.blocks.0.ln_1.bias", "vmunet.layers_up.1.blocks.0.self_attention.x_proj_weight", "vmunet.layers_up.1.blocks.0.self_attention.dt_projs_weight", "vmunet.layers_up.1.blocks.0.self_attention.dt_projs_bias", "vmunet.layers_up.1.blocks.0.self_attention.A_logs", "vmunet.layers_up.1.blocks.0.self_attention.Ds", "vmunet.layers_up.1.blocks.0.self_attention.total_ops", "vmunet.layers_up.1.blocks.0.self_attention.total_params", "vmunet.layers_up.1.blocks.0.self_attention.in_proj.weight", "vmunet.layers_up.1.blocks.0.self_attention.conv2d.weight", "vmunet.layers_up.1.blocks.0.self_attention.conv2d.bias", "vmunet.layers_up.1.blocks.0.self_attention.act.total_ops", "vmunet.layers_up.1.blocks.0.self_attention.act.total_params", "vmunet.layers_up.1.blocks.0.self_attention.out_norm.weight", "vmunet.layers_up.1.blocks.0.self_attention.out_norm.bias", "vmunet.layers_up.1.blocks.0.self_attention.out_proj.weight", "vmunet.layers_up.1.blocks.0.drop_path.total_ops", "vmunet.layers_up.1.blocks.0.drop_path.total_params", "vmunet.layers_up.1.blocks.1.total_ops", "vmunet.layers_up.1.blocks.1.total_params", "vmunet.layers_up.1.blocks.1.ln_1.weight", "vmunet.layers_up.1.blocks.1.ln_1.bias", "vmunet.layers_up.1.blocks.1.self_attention.x_proj_weight", "vmunet.layers_up.1.blocks.1.self_attention.dt_projs_weight", "vmunet.layers_up.1.blocks.1.self_attention.dt_projs_bias", "vmunet.layers_up.1.blocks.1.self_attention.A_logs", "vmunet.layers_up.1.blocks.1.self_attention.Ds", "vmunet.layers_up.1.blocks.1.self_attention.total_ops", "vmunet.layers_up.1.blocks.1.self_attention.total_params", "vmunet.layers_up.1.blocks.1.self_attention.in_proj.weight", "vmunet.layers_up.1.blocks.1.self_attention.conv2d.weight", "vmunet.layers_up.1.blocks.1.self_attention.conv2d.bias", "vmunet.layers_up.1.blocks.1.self_attention.act.total_ops", "vmunet.layers_up.1.blocks.1.self_attention.act.total_params", "vmunet.layers_up.1.blocks.1.self_attention.out_norm.weight", "vmunet.layers_up.1.blocks.1.self_attention.out_norm.bias", "vmunet.layers_up.1.blocks.1.self_attention.out_proj.weight", "vmunet.layers_up.1.blocks.1.drop_path.total_ops", "vmunet.layers_up.1.blocks.1.drop_path.total_params", "vmunet.layers_up.1.upsample.total_ops", "vmunet.layers_up.1.upsample.total_params", "vmunet.layers_up.1.upsample.expand.weight", "vmunet.layers_up.1.upsample.norm.weight", "vmunet.layers_up.1.upsample.norm.bias", "vmunet.layers_up.2.total_ops", "vmunet.layers_up.2.total_params", "vmunet.layers_up.2.blocks.total_ops", "vmunet.layers_up.2.blocks.total_params", "vmunet.layers_up.2.blocks.0.total_ops", "vmunet.layers_up.2.blocks.0.total_params", "vmunet.layers_up.2.blocks.0.ln_1.weight", "vmunet.layers_up.2.blocks.0.ln_1.bias", "vmunet.layers_up.2.blocks.0.self_attention.x_proj_weight", "vmunet.layers_up.2.blocks.0.self_attention.dt_projs_weight", "vmunet.layers_up.2.blocks.0.self_attention.dt_projs_bias", "vmunet.layers_up.2.blocks.0.self_attention.A_logs", "vmunet.layers_up.2.blocks.0.self_attention.Ds", "vmunet.layers_up.2.blocks.0.self_attention.total_ops", "vmunet.layers_up.2.blocks.0.self_attention.total_params", "vmunet.layers_up.2.blocks.0.self_attention.in_proj.weight", "vmunet.layers_up.2.blocks.0.self_attention.conv2d.weight", "vmunet.layers_up.2.blocks.0.self_attention.conv2d.bias", "vmunet.layers_up.2.blocks.0.self_attention.act.total_ops", "vmunet.layers_up.2.blocks.0.self_attention.act.total_params", "vmunet.layers_up.2.blocks.0.self_attention.out_norm.weight", "vmunet.layers_up.2.blocks.0.self_attention.out_norm.bias", "vmunet.layers_up.2.blocks.0.self_attention.out_proj.weight", "vmunet.layers_up.2.blocks.0.drop_path.total_ops", "vmunet.layers_up.2.blocks.0.drop_path.total_params", "vmunet.layers_up.2.blocks.1.total_ops", "vmunet.layers_up.2.blocks.1.total_params", "vmunet.layers_up.2.blocks.1.ln_1.weight", "vmunet.layers_up.2.blocks.1.ln_1.bias", "vmunet.layers_up.2.blocks.1.self_attention.x_proj_weight", "vmunet.layers_up.2.blocks.1.self_attention.dt_projs_weight", "vmunet.layers_up.2.blocks.1.self_attention.dt_projs_bias", "vmunet.layers_up.2.blocks.1.self_attention.A_logs", "vmunet.layers_up.2.blocks.1.self_attention.Ds", "vmunet.layers_up.2.blocks.1.self_attention.total_ops", "vmunet.layers_up.2.blocks.1.self_attention.total_params", "vmunet.layers_up.2.blocks.1.self_attention.in_proj.weight", "vmunet.layers_up.2.blocks.1.self_attention.conv2d.weight", "vmunet.layers_up.2.blocks.1.self_attention.conv2d.bias", "vmunet.layers_up.2.blocks.1.self_attention.act.total_ops", "vmunet.layers_up.2.blocks.1.self_attention.act.total_params", "vmunet.layers_up.2.blocks.1.self_attention.out_norm.weight", "vmunet.layers_up.2.blocks.1.self_attention.out_norm.bias", "vmunet.layers_up.2.blocks.1.self_attention.out_proj.weight", "vmunet.layers_up.2.blocks.1.drop_path.total_ops", "vmunet.layers_up.2.blocks.1.drop_path.total_params", "vmunet.layers_up.2.upsample.total_ops", "vmunet.layers_up.2.upsample.total_params", "vmunet.layers_up.2.upsample.expand.weight", "vmunet.layers_up.2.upsample.norm.weight", "vmunet.layers_up.2.upsample.norm.bias", "vmunet.layers_up.3.total_ops", "vmunet.layers_up.3.total_params", "vmunet.layers_up.3.blocks.total_ops", "vmunet.layers_up.3.blocks.total_params", "vmunet.layers_up.3.blocks.0.total_ops", "vmunet.layers_up.3.blocks.0.total_params", "vmunet.layers_up.3.blocks.0.ln_1.weight", "vmunet.layers_up.3.blocks.0.ln_1.bias", "vmunet.layers_up.3.blocks.0.self_attention.x_proj_weight", "vmunet.layers_up.3.blocks.0.self_attention.dt_projs_weight", "vmunet.layers_up.3.blocks.0.self_attention.dt_projs_bias", "vmunet.layers_up.3.blocks.0.self_attention.A_logs", "vmunet.layers_up.3.blocks.0.self_attention.Ds", "vmunet.layers_up.3.blocks.0.self_attention.total_ops", "vmunet.layers_up.3.blocks.0.self_attention.total_params", "vmunet.layers_up.3.blocks.0.self_attention.in_proj.weight", "vmunet.layers_up.3.blocks.0.self_attention.conv2d.weight", "vmunet.layers_up.3.blocks.0.self_attention.conv2d.bias", "vmunet.layers_up.3.blocks.0.self_attention.act.total_ops", "vmunet.layers_up.3.blocks.0.self_attention.act.total_params", "vmunet.layers_up.3.blocks.0.self_attention.out_norm.weight", "vmunet.layers_up.3.blocks.0.self_attention.out_norm.bias", "vmunet.layers_up.3.blocks.0.self_attention.out_proj.weight", "vmunet.layers_up.3.blocks.0.drop_path.total_ops", "vmunet.layers_up.3.blocks.0.drop_path.total_params", "vmunet.layers_up.3.upsample.total_ops", "vmunet.layers_up.3.upsample.total_params", "vmunet.layers_up.3.upsample.expand.weight", "vmunet.layers_up.3.upsample.norm.weight", "vmunet.layers_up.3.upsample.norm.bias", "vmunet.final_up.total_ops", "vmunet.final_up.total_params", "vmunet.final_up.expand.weight", "vmunet.final_up.norm.weight", "vmunet.final_up.norm.bias", "vmunet.final_conv.weight", "vmunet.final_conv.bias". Unexpected key(s) in state_dict: "model", "optimizer", "lr_scheduler", "max_accuracy", "scaler", "epoch", "config", "model_ema", "max_accuray_ema", "model_ori".
最新发布
05-23
在加载 PyTorch 模型时，`state_dict` 键不匹配问题通常由以下原因引起： 1. **模型结构与保存的权重不一致**：模型定义可能发生了更改，导致保存的权重无法映射到当前模型。 2. **键名不匹配**：保存的权重文件中的...
PyTorch加载模型model.load_state_dict()问题，Unexpected key(s) in state_dict: “module.features..，Expected .
热门推荐
是否龍磊磊真的一无所有的博客
04-17 6万+
希望将训练好的模型加载到新的网络上。如上面题目所描述的，PyTorch在加载之前保存的模型参数的时候，遇到了问题。 Unexpected key(s) in state_dict: "module.features. ...".，Expected ".features....". 直接原因是key值名字不对应。 表明了加载过程中，期望获得的key值为feature...，而不是...
Unexpected key(s) in state_dict: “module.backbone.bn1.num_batches_tracked“
牛肉胡辣汤
11-06 2406
当加载保存的​​时，出现​​错误通常是由于键名不匹配引起的。我们可以通过查看模型的键名和保存的​​的键名来找出不匹配的键，并相应地修改它们。另外，使用​​包装模型可以解决多GPU训练导致的键名前缀问题。希望本文能帮助你解决​​错误，并顺利加载模型参数。
Missing key(s) in state_dict: Unexpected key(s) in state_dict 两种解决方法 祝你成功！！！
weixin_61075104的博客
02-20 2095
Missing key(s) in state_dict: Unexpected key(s) in state_dict
ValueError: (‘Unexpected parameters in params‘, dict_keys([‘cv‘]))
data+scenario+science+insight
10-11 1748
ValueError: ('Unexpected parameters in params', dict_keys(['cv']))
Missing key(s) in state_dict | unexpected_keys | 模型加载 [torch.load] 报错【解决方案】
专注于AI领域前沿技术学习与分享：目标检测、图像修复、超分重建、AI工程化
09-24 1万+
文章目录Missing key(s) in state_dict 报错分析该报错快速直接解决方法额外思路【Unexpected key(s) in state_dict】- 解决思路 Missing key(s) in state_dict 报错分析 在执行 yolov3 pytorch 模型加载时遇到如下报错 类似报错之前其实也遇到过,今天总结一下 所运行代码 # Initialize model model = Darknet(opt.cfg, imgsz) #
yolov8 + EMA 出现的问题：KeyError: ‘EMA‘
qq_45640098的博客
12-12 1323
KeyError: 'EMA'解决方法
Pytorch：Unexpected key(s) in state_dict:
wchzh2015的博客
12-25 1万+
实验环境 pytorch 0.4 Python 3.6 在使用交叉验证后，将保存的模型重新加载求平均的结果。但是出现了以下的错误。 实验的bug截图 在此感谢这篇博文：https://blog.youkuaiyun.com/jacke121/article/details/82495184 自己修改后的关键代码如下： model_resnet101 = get_net...
成功解决 params, model_params = config_reader()config_reader KeyError: ‘param‘
yxh8888888的博客
06-16 671
成功解决 params, model_params = config_reader()config_reader KeyError: ‘param‘
Missing key(s) in state_dict: "log_std". Unexpected key(s) in state_dict: "std".
03-24
<think>嗯，用户遇到了一个关于PyTorch模型状态字典的问题，具体是“Missing key(s) in state_dict: 'log_std'”和“Unexpected key(s) in state_dict: 'std'”。我需要仔细分析这个问题，并给出解决方案。 首先，这个问题通常发生在加载预训练模型时，模型的参数名称不匹配。用户可能在保存模型的时候使用了不同的参数名，导致加载时找不到对应的键。例如，保存的模型有一个参数叫'std'，但加载时模型期望的是'log_std'。这时候，PyTorch就会提示缺少'log_std'，而多出了'std'。 接下来，我需要考虑可能的原因。最常见的原因是模型结构在保存和加载时不一致。可能用户在训练时使用了自定义的层或参数名，而在加载时模型结构发生了变化，或者加载模型的代码没有正确对应保存时的结构。此外，用户可能在保存模型时修改了参数名称，例如将'log_std'保存为'std'，或者相反。 然后，解决方案部分。首先，应该检查模型的结构，确保保存和加载时的模型定义一致。如果参数只是名称不同，可以通过重命名键来解决。例如，将'std'改为'log_std'。这可以通过遍历state_dict并修改键名来实现。另外，如果用户使用的是别人的预训练模型，可能需要查看文档，确认是否有参数名的变化。最后，在加载模型时，使用strict=False参数可以忽略不匹配的键，但需要注意这可能导致模型性能问题，因为部分参数未被加载。 我还需要考虑用户可能的背景。他们可能对PyTorch有一定了解，但在处理模型保存和加载时遇到了困难。可能需要具体的代码示例来帮助理解如何重命名键。同时，提醒用户注意使用strict=False的风险，避免潜在的错误。 最后，总结问题，强调检查模型结构一致性和参数名匹配的重要性，并提供两种解决方案：修改键名或使用非严格加载。确保回答清晰，步骤明确，帮助用户逐步解决问题。</think>这个问题通常出现在PyTorch模型参数加载过程中，涉及模型`state_dict`的键名不匹配。以下是逐步分析及解决方案： --- ### **问题本质** ```python Missing key(s) in state_dict: "log_std" Unexpected key(s) in state_dict: "std" ``` - **直接原因**：加载的模型参数文件（`state_dict`）中存在名为`std`的参数，但当前模型期望的参数名为`log_std`。 - **技术背景**：PyTorch通过`state_dict`的键名严格匹配参数。若模型定义（`__init__()`）与参数文件中的键名不一致，就会触发此错误。 --- ### **常见场景** 1. **自定义模型改动** 修改了模型的参数命名（如将`log_std`重命名为`std`），但未同步更新预训练参数文件。 2. **第三方模型加载** 尝试加载他人训练的模型时，对方模型定义与你本地代码不一致。 3. **版本兼容性问题** 框架或库版本更新导致参数命名规则变化。 --- ### **解决方案** #### 方法1：重命名键名（推荐） 修改`state_dict`中的键名以匹配当前模型定义： ```python import torch # 加载原始参数文件 checkpoint = torch.load("model.pth") # 重命名键：将 "std" 改为 "log_std" new_checkpoint = {} for key, value in checkpoint.items(): new_key = key.replace("std", "log_std") new_checkpoint[new_key] = value # 加载修改后的参数到模型 model.load_state_dict(new_checkpoint, strict=True) ``` #### 方法2：非严格模式加载 允许忽略不匹配的键名（需谨慎）： ```python model.load_state_dict(checkpoint, strict=False) ``` - ✅ 优点：快速绕过错误 - ❌ 缺点：可能导致部分参数未加载，影响模型性能 --- ### **验证步骤** 1. **检查模型定义** 确认模型中是否存在`log_std`参数（如策略梯度算法中的对数标准差）。 2. **对比键名列表** ```python print("Model expects keys:", model.state_dict().keys()) print("Checkpoint has keys:", checkpoint.keys()) ``` 3. **参数映射分析** 若存在多个不匹配键名，需系统性修复（如批量替换`old_key`→`new_key`）。 --- ### **预防措施** 1. **保持命名一致性** 在模型训练和加载过程中使用相同的参数命名规范。 2. **版本控制** 对模型定义和参数文件进行版本标记（如Git标签）。 3. **导出参数前验证** ```python torch.save(model.state_dict(), "model.pth") # 保存前确保参数名正确 ``` --- ### **扩展知识** - `state_dict`是PyTorch模型的核心属性，存储所有可学习参数（权重、偏置等）的键值对。 - 若遇到更复杂的键名不匹配问题，可使用`collections.OrderedDict`重构参数字典。
评论 1
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值