rian-kh / discordmessages-ai Goto Github PK

The compiled program using PyInstaller can't use the subprocess.Popen("python script.py") because the python interpreter isn't included. It can however use exec() but this leads to issues with how to save/terminate it, and subprocess doesn't seem to be usuable to call only functions like exec() (..?).

Tried switching from using subprocess to multiprocessing (branch: multiprocess):

• Program can run the script, but there's no way to terminate it because the process variable (p) can't be referenced outside of runTraining() (p is not defined)

What should happen:

1. Pressing "Start training" starts a thread that runs the process of running trainGPT2.py through exec(). (Works)
2. Pressing "Save and end training" sends a SIGINT signal to the process, forcing it to save. (Doesn't work)

All of the training step output and the text generation output is missing when ran from a console/terminal using the "python" command.

Probably has to do with these tasks being separate python processes (?) that print to themselves rather than the main GUI python process.

Fixing this would make it so that you can run main.py from console, instead of having to set up an IDE.

Main problem is figuring out how to terminate RIGHT AFTER saving.

In this code for a combined "End training" button:

if event == 'End training' and trainingStarted:
    p.send_signal(signal.SIGINT)
    p.terminate()
    trainingStarted = False

Because it's a threaded task (?) it will try saving, but immediately exit before it can fully save.
Sending the SIGINT (needed for it to save) only interrupts the process but does not stop it from running in the background, so the terminate is necessary.

Adding the line:

window.find_element('_output_').Update(disabled=True)

to the event loop hangs the program once closed with training/generation running.

When adding this line, the output box can be typed in when the program is first started but is disabled as it should be once a button is pressed.

Would copy site-packages to python environment's executable folder to update the gpt_2.py file

Mainly because model deletion relies on booleans updated when pressing their respective "End" buttons, but in the case where the program exits by itself (steps reached, samples finished) the boolean is not updated.

• Could use thread/genThread.wait(), but this would result in the program hanging until finished...

• Ignore the check for 'saveRan' for this specific case somehow

Editing the close event:

    if event in (None, "Exit"):

        if trainingStarted:
            sg.popup("Please save and end training before quitting the program.")
            window.refresh()
        else:
            if generationStarted:
                q.terminate()

            exit()
            break

makes it so the popup shows once the app is closed, but the program still exits regardless.

ex. No strings in argument textboxes needing integers, like Steps

Main problem:

• Adding an 'if' block to check if a parameter's input is valid ONLY gets ran if a button is pressed, not if the user is done entering text.

In order, what should happen is:

1. User enters invalid text
2. Once clicking away, a popup says to check again that text is valid
3. Parameter is updated back to default value

Error message when pressing "Save and end training":

Traceback (most recent call last):
File "C:\Users\Rian\PycharmProjects\DiscordMessages-AI\main.py", line 218, in
p.send_signal(signal.SIGINT)
File "C:\Users\Rian\miniconda3\envs\tf\lib\subprocess.py", line 1545, in send_signal
raise ValueError("Unsupported signal: {}".format(sig))
ValueError: Unsupported signal: 2

When pressing "End text generation", no error is thrown but the generation still continues.

Mainly a problem for large sample lengths as it can run in the background for a while, still turns off on terminate (program exit) though

• Make it so only certain buttons (Start/End training, Start/End generation) clear the output box rather than every button.