Software Engineering Courses (SECourses)•3y ago

Sir, I was able to train 2 subjects yesterday with 4500 epochs. I also used at least 1000 class ima

Sir, I was able to train 2 subjects yesterday with 4500 epochs. I also used at least 1000 class images. The results were not very satisfactory. Is it beneficial if I continue training?

CColored Tonalities not if you get theoretical math, if I understand it correctly 🙂

Furkan Gözükara SECourses•1/24/23, 11:05 AM

there is no true random very correct

SSeagreen Sir, I was able to train 2 subjects yesterday with 4500 epochs. I also used at ...

Furkan Gözükara SECourses•1/24/23, 11:05 AM

nope

Furkan Gözükara SECourses•1/24/23, 11:05 AM

more training usually dont get improved

Colored Tonalities•1/24/23, 11:06 AM

my model was at 22.600 and my hypothesis is that you could theoretically train to infinite given the right tokens

Colored Tonalities•1/24/23, 11:06 AM

75 tokens have a lower weight than 1 token

Colored Tonalities•1/24/23, 11:06 AM

compounded

CColored Tonalities 75 tokens have a lower weight than 1 token

Furkan Gözükara SECourses•1/24/23, 11:06 AM

definitely

Furkan Gözükara SECourses•1/24/23, 11:06 AM

this thing has

Furkan Gözükara SECourses•1/24/23, 11:06 AM

between relationships as well

Furkan Gözükara SECourses•1/24/23, 11:06 AM

that is not something

Furkan Gözükara SECourses•1/24/23, 11:06 AM

that can be understood

Furkan Gözükara SECourses•1/24/23, 11:06 AM

manually

Furkan Gözükara SECourses•1/24/23, 11:07 AM

it is all about machine learning

Furkan Gözükara SECourses•1/24/23, 11:07 AM

bigger scheme it works

Colored Tonalities•1/24/23, 11:07 AM

yes

Furkan Gözükara SECourses•1/24/23, 11:07 AM

when you zoom in

Furkan Gözükara SECourses•1/24/23, 11:07 AM

all you see is bunch of numbers

Colored Tonalities•1/24/23, 11:07 AM

yeah

Furkan Gözükara SECourses•1/24/23, 11:07 AM

so when you use more prompts during training

Furkan Gözükara SECourses•1/24/23, 11:07 AM

more relationships are

Furkan Gözükara SECourses•1/24/23, 11:07 AM

compounded

Colored Tonalities•1/24/23, 11:07 AM

How many good images do you have of yourself Dr.Furkan, that are different?

FFurkan Gözükara SECourses nope

SeagreenOP•1/24/23, 11:07 AM

Thank you

Colored Tonalities•1/24/23, 11:08 AM

like "epic" pictures of yourself

Colored Tonalities•1/24/23, 11:08 AM

Ok, but output

Furkan Gözükara SECourses•1/24/23, 11:08 AM

ye just understood

Furkan Gözükara SECourses•1/24/23, 11:08 AM

i got quite good

Furkan Gözükara SECourses•1/24/23, 11:08 AM

afk for a while

Colored Tonalities•1/24/23, 11:11 AM

ok

So pick 300 quite good images that are unique and different.
create a model where you use the original 12 images that you used
12 images PLUS the 300 images WITH filewords.

This should not overtrain the model and because of compounded weight it should DECREASE the variation of the loss/average learning rate

Colored Tonalities•1/24/23, 11:11 AM

You compound the most at 75 Tokens

Colored Tonalities•1/24/23, 11:11 AM

So theoretically if you scan 300 images with 1 Token and 300 images with 75 tokens the Loss average/ learning rate should go down

Colored Tonalities•1/24/23, 11:12 AM

This I will test now

Colored Tonalities•1/24/23, 11:12 AM

Because actually I think there is replicated results even though there might be some standard deviation currently in the process

FFurkan Gözükara SECourses thanks

Nandux•1/24/23, 11:21 AM

Hi again, i followed the instructions of the video. I found them easy although i dont know anything about code. Now, i am not getting the black samples during training but only the sanity sample is related to the subject. The prompt sample is random

Colored Tonalities•1/24/23, 11:21 AM

black samples?

Colored Tonalities•1/24/23, 11:21 AM

what model are you running?

Colored Tonalities•1/24/23, 11:22 AM

Nandux•1/24/23, 11:22 AM

venv "F:\Automatic1111 Webui\stable-diffusion-webui\venv\Scripts\Python.exe"
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Commit hash: 602a1864b05075ca4283986e6f5c7d5bce864e11
Installing requirements for Web UI

#######################################################################################################
Initializing Dreambooth
If submitting an issue on github, please provide the below text for debugging purposes:

Python revision: 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Dreambooth revision: c544ee11aee0085a7fbb7fdda65898dea2145f0c
SD-WebUI revision: 602a1864b05075ca4283986e6f5c7d5bce864e11

Checking Dreambooth requirements...
[+] bitsandbytes version 0.35.0 installed.
[+] diffusers version 0.10.2 installed.
[+] transformers version 4.25.1 installed.
[+] xformers version 0.0.14.dev0 installed.
[+] torch version 1.13.1+cu116 installed.
[+] torchvision version 0.14.1+cu116 installed.

#######################################################################################################

Launching Web UI with arguments: --xformers
=================================================================================
You are running xformers 0.0.14.dev.
The program is tested to work with xformers 0.0.16rc425.
To reinstall the desired version, run with commandline flag --reinstall-xformers.

Colored Tonalities•1/24/23, 11:22 AM

Colored Tonalities•1/24/23, 11:23 AM

Did you do it like this?

CColored Tonalities Click to see attachment

Nandux•1/24/23, 11:23 AM

i am not having the black samples since i followed the instructions of the vid. But the main samples (prompt) seems random stuff. The sanity prompt looks correct

CColored Tonalities Click to see attachment

Nandux•1/24/23, 11:23 AM

let me check it

Colored Tonalities•1/24/23, 11:23 AM