search

open-mmlab / mmpretrain

OpenMMLab Pre-training Toolbox and Benchmark
https://mmpretrain.readthedocs.io/en/latest/
Apache License 2.0
3.4k stars 1.06k forks source link

[Bug]RuntimeError: CUDA error: CUBLAS_STATUS_INVALID_VALUE when calling #477

Closed RainFrost1 closed 2 years ago

RainFrost1 commented 3 years ago

推荐使用英语模板 Bug report，以便你的问题帮助更多人。

描述 bug

简单地描述一下遇到了什么 bug When training swintransformer [填写这里]

复现流程

在命令行中执行的详细操作 python tools/train.py configs/swin_transformer/swin_tiny_224_b16x64_300e_imagenet.py

[填写这里]

相关信息

  1. pip list | grep "mmcv\|mmcls\|^torch" 命令的输出 mmcls==0.15.0 mmcv-full==1.3.13 torch==1.9.0
  2. 如果你修改了，或者使用了新的配置文件，请在这里写明 
    [填写这里]
  3. 如果你是在训练过程中遇到的问题，请填写完整的训练日志和报错信息 [填写这里]
  4. 如果你对 mmcls 文件夹下的代码做了其他相关的修改，请在这里写明 [填写这里]

附加内容

任何其他有关该 bug 的信息、截图等 图片

[填写这里]

mzr1996 commented 3 years ago

Hello, did you modify the config file?

canghaiyunfan commented 2 years ago

@mzr1996 I faced the same issue, maybe a bug for pytorch1.8

459737087 commented 2 years ago

I also met this question, maybe a bug for pytorch 1.9. when I ran it in pytorch1.8, this question disappeared!

Ezra-Yu commented 2 years ago

Make sure that the PyTorch version matches the Cuda version, refer to PyTorch installation guide.