pytorch runtime_error ai_generated true

RuntimeError：DataLoader工作进程（pid 12345）pin_memory()：CUDA错误：无效的设备上下文

RuntimeError: DataLoader worker (pid 12345) pin_memory(): CUDA error: invalid device context

ID: pytorch/dataloader-pin-memory-cuda

其他格式: JSON · Markdown 中文 · English
88%修复率
86%置信度
1证据数
2023-02-10首次发现

版本兼容性

版本状态引入弃用备注
torch>=1.6.0 active — — —
CUDA>=11.0 active — — —

根因分析

使用pin_memory=True的DataLoader会生成工作进程，这些进程尝试从fork的子进程中使用CUDA，导致设备上下文无效，因为CUDA在初始化后不支持fork。

English

DataLoader with pin_memory=True spawns workers that attempt to use CUDA from forked processes, causing invalid device context due to CUDA not supporting fork after initialization.

generic

官方文档

https://pytorch.org/docs/stable/data.html#multi-process-data-loading

解决方案

  1. Set multiprocessing start method to 'spawn': torch.multiprocessing.set_start_method('spawn', force=True) before creating DataLoader
  2. Use pin_memory=False in DataLoader and manually move tensors to GPU after loading
  3. Move CUDA initialization after DataLoader creation or use single-process loading with num_workers=0

无效尝试

常见但无效的做法:

  1. 80% 失败

    Workers need CUDA context for pin_memory; hiding devices breaks the purpose.

  2. 90% 失败

    More workers increase chance of CUDA context issues.