系统cudnn和conda环境cudnn冲突

1.解决冲突，删除conda环境cudnn

报错记录，训练时出现这种报错，

te_internal_implERKNS_7backend11VariantPackEP11CUstream_stRNS0_18LayerNormFwdParamsERKNS1_20NormForwardOperationEmb, version libcudnn_cnn_infer.so.8
Could not load library libcudnn_cnn_train.so.8. Error: /usr/local/cuda-12.6/lib64/libcudnn_cnn_train.so.8: undefined symbol: _ZN5cudnn3cnn34layerNormFwd_execute_internal_implERKNS_7backend11VariantPackEP11CUstream_stRNS0_18LayerNormFwdParamsERKNS1_20NormForwardOperationEmb, version libcudnn_cnn_infer.so.8
Could not load library libcudnn_cnn_train.so.8. Error: /usr/local/cuda-12.6/lib64/libcudnn_cnn_train.so.8: undefined symbol: _ZN5cudnn3cnn34layerNormFwd_execute_internal_implERKNS_7backend11VariantPackEP11CUstream_stRNS0_18LayerNormFwdParamsERKNS1_20NormForwardOperationEmb, version libcudnn_cnn_infer.so.8
Could not load library libcudnn_cnn_train.so.8. Error: /usr/local/cuda-12.6/lib64/libcudnn_cnn_train.so.8: undefined symbol: _ZN5cudnn3cnn34layerNormFwd_execute_internal_implERKNS_7backend11VariantPackEP11CUstream_stRNS0_18LayerNormFwdParamsERKNS1_20NormForwardOperationEmb, version libcudnn_cnn_infer.so.8
Could not load library libcudnn_cnn_train.so.8. Error: /usr/local/cuda-12.6/lib64/libcudnn_cnn_train.so.8: undefined symbol: _ZN5cudnn3cnn34layerNormFwd_execute_internal_implERKNS_7backend11VariantPackEP11CUstream_stRNS0_18LayerNormFwdParamsERKNS1_20NormForwardOperationEmb, version libcudnn_cnn_infer.so.8
Could not load library libcudnn_cnn_train.so.8. Error: /usr/local/cuda-12.6/lib64/libcudnn_cnn_train.so.8: undefined symbol: _ZN5cudnn3cnn34layerNormFwd_execute_internal_implERKNS_7backend11VariantPackEP11CUstream_stRNS0_18LayerNormFwdParamsERKNS1_20NormForwardOperationEmb, version libcudnn_cnn_infer.so.8
Could not load library libcudnn_cnn_train.so.8. Error: /usr/local/cuda-12.6/lib64/libcudnn_cnn_train.so.8: undefined symbol: _ZN5cudnn3cnn34layerNormFwd_execute_internal_implERKNS_7backend11VariantPackEP11CUstream_stRNS0_18LayerNormFwdParamsERKNS1_20NormForwardOperationEmb, version libcudnn_cnn_infer.so.8
Could not load library libcudnn_cnn_train.so.8. Error: /usr/local/cuda-12.6/lib64/libcudnn_cnn_train.so.8: undefined symbol: _ZN5cudnn3cnn34layerNormFwd_execute_internal_implERKNS_7backend11VariantPackEP11CUstream_stRNS0_18LayerNormFwdParamsERKNS1_20NormForwardOperationEmb, version libcudnn_cnn_infer.so.8
Could not load library libcudnn_cnn_train.so.8. Error: /usr/local/cuda-12.6/lib64/libcudnn_cnn_train.so.8: undefined symbol: _ZN5cudnn3cnn34layerNormFwd_execute_internal_implERKNS_7backend11VariantPackEP11CUstream_stRNS0_18LayerNormFwdParamsERKNS1_20NormForwardOperationEmb, version libcudnn_cnn_infer.so.8
Could not load library libcudnn_cnn_train.so.8. Error: /usr/local/cuda-12.6/lib64/libcudnn_cnn_train.so.8: undefined symbol: _ZN5cudnn3cnn34layerNormFwd_execute_internal_implERKNS_7backend11VariantPackEP11CUstream_stRNS0_18LayerNormFwdParamsERKNS1_20NormForwardOperationEmb, version libcudnn_cnn_infer.so.8
Could not load library libcudnn_cnn_train.so.8. Error: /usr/local/cuda-12.6/lib64/libcudnn_cnn_train.so.8: undefined symbol: _ZN5cudnn3cnn34layerNormFwd_execute_internal_implERKNS_7backend11VariantPackEP11CUstream_stRNS0_18LayerNormFwdParamsERKNS1_20NormForwardOperationEmb, version libcudnn_cnn_infer.so.8
Could not load library libcudnn_cnn_train.so.8. Error: /usr/local/cuda-12.6/lib64/libcudnn_cnn_train.so.8: undefined symbol: _ZN5cudnn3cnn34layerNormFwd_execute_internal_implERKNS_7backend11VariantPackEP11CUstream_stRNS0_18LayerNormFwdParamsERKNS1_20NormForwardOperationEmb, version libcudnn_cnn_infer.so.8
Could not load library libcudnn_cnn_train.so.8. Error: /usr/local/cuda-12.6/lib64/libcudnn_cnn_train.so.8: undefined symbol: _ZN5cudnn3cnn34layerNormFwd_execute_internal_implERKNS_7backend11VariantPackEP11CUstream_stRNS0_18LayerNormFwdParamsERKNS1_20NormForwardOperationEmb, version libcudnn_cnn_infer.so.8
Could not load library libcudnn_cnn_train.so.8. Error: /usr/local/cuda-12.6/lib64/libcudnn_cnn_train.so.8: undefined symbol: _ZN5cudnn3cnn34layerNormFwd_execute_internal_implERKNS_7backend11VariantPackEP11CUstream_stRNS0_18LayerNormFwdParamsERKNS1_20NormForwardOperationEmb, version libcudnn_cnn_infer.so.80%|          | 0/200 [00:01<?, ?it/s]
Traceback (most recent call last):File "/home/ubuntu/workdir/xueqin/yolov13/train.py", line 6, in <module>results = model.train(^^^^^^^^^^^^File "/home/ubuntu/workdir/xueqin/yolov13/ultralytics/engine/model.py", line 808, in trainself.trainer.train()File "/home/ubuntu/workdir/xueqin/yolov13/ultralytics/engine/trainer.py", line 207, in trainself._do_train(world_size)File "/home/ubuntu/workdir/xueqin/yolov13/ultralytics/engine/trainer.py", line 389, in _do_trainself.scaler.scale(self.loss).backward()File "/home/ubuntu/miniconda3/envs/yolov13/lib/python3.11/site-packages/torch/_tensor.py", line 522, in backwardtorch.autograd.backward(File "/home/ubuntu/miniconda3/envs/yolov13/lib/python3.11/site-packages/torch/autograd/__init__.py", line 266, in backwardVariable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
RuntimeError: GET was unable to find an engine to execute this computation

具体原因是安装cuda的时候，按照教程将cudmn也安装到了系统cuda路径中，然后在pyhon虚拟环境中使用pip instal torch时会自动安装一个cudnn，两个发生了冲突。

解决办法，卸载conda环境中的cudnn

 pip uninstall nvidia-cudnn-cu12

2.修改终端cuda

nano ~/.bashrc

• 作用：用 nano 文本编辑器打开当前用户的 ~/.bashrc 文件。

• ~/.bashrc 是一个 Bash Shell 的启动配置文件，每次你打开一个新的 终端（非登录 shell） 时，它就会被执行。

运行nano ~/.bashrc，拉到文本最下面就可以修改了。

退出时Ctrl + O （保存）→ 回车 → Ctrl + X（退出）

source ~/.bashrc

让刚修改的 ~/.bashrc 立即生效。