Win10环境下30系显卡安装TensorFlow1.15

众所周知,TensorFlow1.x并不支持30系显卡,需要进行另外的设置。

  1. Win10环境配置好最新的NVIDIA环境,如cuda和cudnn;

  2. 具体代码如下:

    conda create -n tf115rtx3090ti python==3.7
    conda activate tf115rtx3090ti
    conda install cudnn==8.2.1
    pip install keras==2.3.1
    pip install tensorflow-1.15.4+nv-cp37-cp37m-win_amd64.whl

此时,可能仍然无法正常使用TensorFlow,需要按照错误提示,重新安装对应版本pip install protobuf

原链接教程中,有提到修改anaconda3\envs\tf115RTX\lib\site-packages\keras\engine\saving.py,将其中.decode('utf8')删除。实际测试中,未删除依然能够成功。

参考链接:
https://blog.csdn.net/Straka/article/details/118851912?utm_medium=distribute.pc_relevant.none-task-blog-2~default~baidujs_title~default-0-118851912-blog-119059206.pc_relevant_multi_platform_whitelistv3&spm=1001.2101.3001.4242.1&utm_relevant_index=3
https://github.com/Fannhhyy/tensorflow1.15-whl-and-cpp-api-for-win-and-rtx3090
https://blog.csdn.net/qq_43559448/article/details/86698385
https://zhuanlan.zhihu.com/p/356526953
https://blog.csdn.net/weixin_36896856/article/details/121463575
https://blog.csdn.net/u013841196/article/details/119059206

本文链接:

https://ma.ge/archives/622.html