South China Technology stack cnn+bilstm+attention

The present situation

already installed cuda cudnn anaconda pycharm

cuda 9.0 Corresponding cudnn Is already installed , Corresponding python yes 3.7 And the following

Tried anaconda Add source Creating a new environment cannot

The goal is ： To configure Win10 Next Of pytorch
The following is my two-day effort and error reporting ：
Hope to get advice ！！

Basic introduction
bilstm Introduce ：（ Use wechat to login ）

Implementation code
attention Introduce
bert layer （ Most come From the cbow）
Do the preliminary training ,
A very good one bert Explain

The blogger bert+cnn And to Code
Include attention Mechanism ,
In short, the blogger is a treasure blogger ,
txt and CSV The difference between ,
（ Not very clear ..）
Yes txt Divide the data

In training It seems to use cuda

I installed it myself 9.1 Version of . Corresponding python yes 3.7 And the following .

Then verify It's installed .

It still won't work after installation .
Find a no GPU Methods
Of course, I definitely want to use ,, Try it first and then debug
So I'll look for it the next day , Be careful Version correspondence .
acutely .. Graphics card reason . Lowest cuda All are 10.2... Out of line , Or use it CPU Try it .

Pure despises 9.1 My people WOW ..

Finally take anaconda Route to solve environmental problems

Just pretend it's over Actually, it works well （ This benefit comes with it Learn from Li Mu's Notepad in depth , It should work ）
Then the perfect construction plan It should be this , The steps are reliable
then Unload and reload cuda
This is reinstall A more detailed tutorial

pycharm install torch and cuda（ stay anaconda Create a new environment ）

cudnn Download requires a password , Account email 89, password Szy73041
This is my cuda The version corresponds to cudnn

After downloading, unzip it and put it in the corresponding folder . Copy them into the folder respectively , It is equivalent to a supplementary function .
verification ：

win+R,cmd Open the command window , Input ： cd /d C:\Program Files\NVIDIA GPU Computing
Toolkit\CUDA\v11.3\extras\demo_suite Enter to switch the path Input ：bandwidthTest.exe
Show PASS Installation succeeded Input ：deviceQuery.exe Show PASS, be cuda and cudnn All installed successfully .

Two pass.... Real tears ,,, The installation should have been successful ,

This article A perfect explanation anaconda Import environment to pycharm
It's mainly about looking at anaconda To configure python Directory location , But I don't think it needs so much trouble, as if .
I'll see if I can find it in his last catalogue .
So the next idea is anaconda In the configuration python Interpreter , Then import to pycharm.
,
anaconda Need to be in a Update the source in the configuration file ,

After the virtual environment creation fails and an error is reported , The last stubbornness

(base) C:\Users\Administrator>conda create -n pytorch python=3.6
Collecting package metadata (current_repodata.json): failed

ProxyError: Conda cannot proceed due to an error in your proxy
configuration. Check for typos and other configuration errors in any
‘.netrc’ file in your home directory, any environment variables ending
in ‘_PROXY’, and any other system-wide proxy configuration settings.

Too tired ... such .（ I'm thinking about changing it again CPU 了 ）