查看当前环境中的库

conda env list

卸载

pip uninstall

pip uninstall tensorflow               
pip uninstall tensorflow-intel       
pip uninstall tensorflow-io-gcs-filesystem

由于之前已经下载了cuda

只需要下载cuDNNcuDNN 9.4.0 Downloads | NVIDIA Developer

cudda和cuda对应网址

从源代码构建  |  TensorFlow (google.cn)

解压

打开cudnn中文件

放在CUDA目录下

可以直接替换

使用镜像源安装

pip install tensorflow-gpu==2.6.0 -i https://pypi.tuna.tsinghua.edu.cn/simple

测试，输入

python

输入

import tensorflow as tf

报错

If you cannot immediately regenerate your protos, some other possible workarounds are:
 1. Downgrade the protobuf package to 3.20.x or lower.
 2. Set PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python (but this will use pure-Python parsing and will be much slower).

输入exit()退出python环境

pip uninstall protobuf

pip install protobuf==3.19.0

报错

是numpy版本不符合

pip install numpy==1.21.6 -i https://pypi.tuna.tsinghua.edu.cn/simple/

再进入python环境测试成功

进入pycharm报错

cannot import name 'dtensor' from 'tensorflow.compat.v2.experimental'

重新更新keras版本

pip install --upgrade keras==2.6.0

报错ran out of memory (OOM)

前面加

import os
os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
import tensorflow as tf

用cpu运行