#1388 AttributeError: type object 'torch._C._distributed_c10d.BackendType' has no attribute 'XCCL'

这个错误表明 PyTorch 在尝试初始化 CUDA 时遇到了问题，具体是分布式计算模块中的 BackendType.XCCL 属性缺失。这通常与 PyTorch 版本、CUDA 驱动版本或显卡硬件兼容性有关。

请按以下步骤排查：

1. 临时禁用 CUDA 加速：

   • 在软件主界面，取消勾选 "CUDA加速" 选项。
   • 这将强制软件使用 CPU 进行语音识别，虽然速度较慢，但可以绕过当前的 GPU 兼容性问题。

2. 检查显卡和驱动：

   • 确认您的显卡是 NVIDIA 显卡。AMD 或 Intel 集成显卡不支持 CUDA。
   • 前往 NVIDIA 官网更新您的显卡驱动至最新版本。

3. 检查 CUDA 工具包：

   • 预打包版的 pyVideoTrans 已内置了特定版本的 CUDA 运行时。此错误可能意味着该内置版本与您的系统环境存在冲突。
   • 如果您系统上安装了其他版本的 CUDA，可以尝试卸载它们，仅使用软件内置的版本。

如果以上步骤无法解决问题，或者您需要继续使用 GPU 加速，请提供 E:/Pyvideotrans视频翻译软件/videotrans/logs/ 目录下的最新日志文件，以便进行更深入的分析。

from torch import _VF, Tensor

File "", line 1027, in _find_and_load
File "", line 1006, in _find_and_load_unlocked
File "", line 688, in _load_unlocked
File "pyimod02_importers.py", line 457, in exec_module
File "torch\nn\__init__.py", line 8, in

from torch

......
, in

from torch import _VF, Tensor

File "", line 1027, in _find_and_load
File "", line 1006, in _find_and_load_unlocked
File "", line 688, in _load_unlocked
File "pyimod02_importers.py", line 457, in exec_module
File "torch\nn\__init__.py", line 8, in

from torch.nn.modules import *  # usort: skip # noqa: F403

File "", line 1027, in _find_and_load
File "", line 1006, in _find_and_load_unlocked
File "", line 688, in _load_unlocked
File "pyimod02_importers.py", line 457, in exec_module
File "torch\nn\modules\__init__.py", line 1, in

from .module import Module  # usort: skip

File "", line 1027, in _find_and_load
File "", line 1006, in _find_and_load_unlocked
File "", line 688, in _load_unlocked
File "pyimod02_importers.py", line 457, in exec_module
File "torch\nn\modules\module.py", line 17, in

overload,

File "", line 1027, in _find_and_load
File "", line 1006, in _find_and_load_unlocked
File "", line 688, in _load_unlocked
File "pyimod02_importers.py", line 457, in exec_module
File "torch\utils\__init__.py", line 8, in

from torch.utils import (

File "", line 1027, in _find_and_load
File "", line 1006, in _find_and_load_unlocked
File "", line 688, in _load_unlocked
File "pyimod02_importers.py", line 457, in exec_module
File "torch\utils\data\__init__.py", line 1, in

from torch.utils.data.dataloader import (

File "", line 1027, in _find_and_load
File "", line 1006, in _find_and_load_unlocked
File "", line 688, in _load_unlocked
File "pyimod02_importers.py", line 457, in exec_module
File "torch\utils\data\dataloader.py", line 21, in

import torch.utils.data.graph_settings

File "", line 1027, in _find_and_load
File "", line 1006, in _find_and_load_unlocked
File "", line 688, in _load_unlocked
File "pyimod02_importers.py", line 457, in exec_module
File "torch\distributed\__init__.py", line 122, in

from .device_mesh import DeviceMesh, init_device_mesh

File "", line 1027, in _find_and_load
File "", line 1006, in _find_and_load_unlocked
File "", line 688, in _load_unlocked
File "pyimod02_importers.py", line 457, in exec_module
File "torch\distributed\device_mesh.py", line 40, in

_find_pg_by_ranks_and_tag,

File "", line 1027, in _find_and_load
File "", line 1006, in _find_and_load_unlocked
File "", line 688, in _load_unlocked
File "pyimod02_importers.py", line 457, in exec_module
File "torch\distributed\distributed_c10d.py", line 236, in

initial value of some fields. Users should neither use it directly

File "torch\distributed\distributed_c10d.py", line 287, in Backend

devices: Optional[Union[str, List[str]]] = None,

AttributeError: type object 'torch._C._distributed_c10d.BackendType' has no attribute 'XCCL'
等待 uuid_signal 线程退出
正在等待线程 PrepareVideo 结束...
Attempting to kill ffmpeg processes for user: Hiker
Successfully killed ffmpeg processes using taskkill

这个错误是 PyTorch 分布式计算模块的兼容性问题，具体是 BackendType.XCCL 属性缺失。这通常发生在特定版本的 PyTorch 与系统环境不兼容时。

立即解决方案：

1. 禁用 CUDA 加速：在软件主界面，务必取消勾选 "CUDA加速" 选项。这将强制软件使用 CPU 进行语音识别，可以绕过当前的 GPU 兼容性问题。

如果仍需使用 GPU 加速，请尝试：

2. 更新显卡驱动：前往 NVIDIA 官网下载并安装最新的显卡驱动程序。
3. 检查软件完整性：重新下载完整的软件包（约 2.5G），确保所有文件完整无损坏。
4. 系统兼容性：确认您的 Windows 10 系统已更新到最新版本。

如果以上方法均无效，建议暂时使用 CPU 模式进行处理。虽然速度较慢，但功能完全正常。这个错误与软件内置的 PyTorch 版本有关，可能需要等待软件更新来解决底层依赖的兼容性问题。