Cant reproduce extra many files were generated while executing "each operation times usage"

General Discussion
1

While executing " each operation times usage" each operation times usage | Khadas Documentation
I got many files in the folder op_test\bin_r_cv4 on khadas VIM3.
My question is: what do these files mean?
I cannot regenerate these files again. I am following the same steps but I cannot generate these files again. I don’t know which steps I followed when these files were generated.

Screenshot 2022-03-30 045249
Screenshot 2022-03-30 045249714×717 168 KB

the folder is at this link.

@numbqq @Frank

2

I think I followed the same steps as I am following now. But now I can not generate these files. Can you please help me out?

3

@enggsajjad Each file represents a neural network layer.

This is reproducible, I suggest you can re-run it exactly as documented

4

I tried many times, all the steps are executed normally, but there is no text files produced. This is the output of all the steps, but there is no text files produced as in the previous step.

5

@enggsajjad I will find time to test, but you need to wait

6

Oh, I just figured out that while setting the variables using export command. I was using Ctrl+R to use previously used commands in the terminal and I guess I missed

export NN_LAYER_DUMP=1

and typed

export CNN_PERF=1

twice. I will check it today and update here.
btw, where i can find the documentation of these export variables?

7

Sorry, what does the numbers in these files mean? Where can I get the information about these files and numbers? @Frank @numbqq

8

@enggsajjad Have you checked your network model before conversion? The network model is numbered from the top to the bottom. This number corresponds to the order of the operators of the neural network layer.

9

Sorry, I am talking about the contents of the files produced using export NN_LAYER_DUMP=1. as below

image
image271×809 14.1 KB

Thanks @Frank for quick response.

Also, I am really new to NPU and model conversions, I have not checked the network model before conversion.

10

I just explored these files and found that there is a link between number of enteries in these text files in the folder op_test[bin_r_cv4](Dropbox - bin_r_cv4 - Simplify your life) and corresponding layer output in graph.json. For example:
The files:
image
contains following Total numder of enteries:
image

So I guess these are the weights or so? Please guide?
Also do you have any idea how to interpret or understand the graph.json, please recommend any documentation,
Thanks @Frank @numbqq
Regards,