Software Engineering Courses (SECourses)•3y ago

I just realised, that I installed Kohya_ss choosing Torch 2.0, and you mentioned in the video that C

I just realised, that I installed Kohya_ss choosing Torch 2.0, and you mentioned in the video that CuDNN is not necessary with Torch 2.0. So that installing it is not necessary. So case closed. Good thing I watch your LoRA video to catch that detail. Thanks again

hopeyan•4/30/23, 3:38 PM

https://github.com/ddPn08/Radiata

GitHub

GitHub - ddPn08/Radiata: Stable diffusion webui based on diffusers.

Stable diffusion webui based on diffusers. Contribute to ddPn08/Radiata development by creating an account on GitHub.

hopeyan•4/30/23, 3:38 PM

@Furkan Gözükara SECourses

Zono50•4/30/23, 4:48 PM

@hopeyan does it run better than automatic1111?

FFurkan Gözükara SECourses thanks

rafrafOP•4/30/23, 4:50 PM

You still like webui dreambooth better than koya/Lora? Your video on Kohya Lora looks really good

Zono50•4/30/23, 4:52 PM

deepfloyd has to cut down on it's vram usage, 16gb is kinda ridiculous for making some text

FFurkan Gözükara SECourses negative tokens are used to derive away image from those latent space. they are ...

Михаил Кислов•4/30/23, 5:01 PM

It turns out that you can enter an arbitrarily long request into a negative prompt and its accuracy will not decrease?

Rrafraf I just realised, that I installed Kohya_ss choosing Torch 2.0, and you mentioned...

Furkan Gözükara SECourses•4/30/23, 5:43 PM

you are welcome. yes torch 2 has 8.7 dlls

Hhopeyan https://github.com/ddPn08/Radiata

Furkan Gözükara SECourses•4/30/23, 5:44 PM

very nice supports DeepFloyd IF

Furkan Gözükara SECourses•4/30/23, 5:44 PM

i should make a tutorial for this

Rrafraf You still like webui dreambooth better than koya/Lora? Your video on Kohya Lora ...

Furkan Gözükara SECourses•4/30/23, 5:44 PM

actually i am comparing atm :d

Furkan Gözükara SECourses•4/30/23, 5:44 PM

doing rank 128 192 and 256 lora training

Furkan Gözükara SECourses•4/30/23, 5:44 PM

i have done dreambooth

Furkan Gözükara SECourses•4/30/23, 5:44 PM

will compare

ZZono50 deepfloyd has to cut down on it's vram usage, 16gb is kinda ridiculous for makin...

Furkan Gözükara SECourses•4/30/23, 5:45 PM

ye even lowest is high

Furkan Gözükara SECourses•4/30/23, 5:45 PM

but in video i used kaggle

Furkan Gözükara SECourses•4/30/23, 5:45 PM

so everyone will be able to use

ММихаил Кислов It turns out that you can enter an arbitrarily long request into a negative prom...

Furkan Gözükara SECourses•4/30/23, 5:45 PM

no it will. negative prompts also do affect your final results significantly

dave5555•4/30/23, 7:11 PM

dr. F did u ever figure out the batch loader for control nets?

dave5555•4/30/23, 7:12 PM

as in dif batches for each control net seperately

Ddave5555 as in dif batches for each control net seperately

Furkan Gözükara SECourses•4/30/23, 7:14 PM

yes there was a pull request and it had worked for me,

Furkan Gözükara SECourses•4/30/23, 7:24 PM

wow lora super memorized the background and clothing interestingly

Furkan Gözükara SECourses•4/30/23, 7:25 PM

i will test with captions

Furkan Gözükara SECourses•4/30/23, 7:31 PM

dreambooth had much better flexibility

Zono50•4/30/23, 7:45 PM

did u upload the video yet? never heard of kaggle before

Zono50•4/30/23, 7:47 PM

yeah i've had much better success with captions, without, my model was burn after 2 epochs

ZZono50 did u upload the video yet? never heard of kaggle before

Furkan Gözükara SECourses•4/30/23, 7:49 PM

still processing

Furkan Gözükara SECourses•4/30/23, 7:49 PM

kaggle is even better than colab

ZZono50 yeah i've had much better success with captions, without, my model was burn afte...

Furkan Gözükara SECourses•4/30/23, 7:49 PM

ye i am testing git captions

Furkan Gözükara SECourses•4/30/23, 7:53 PM

here lora vs dreambooth

Zono50•4/30/23, 7:54 PM

i usually do mine like this

Furkan Gözükara SECourses•4/30/23, 7:54 PM

i will show in a minute

Zono50•4/30/23, 7:54 PM

Zono50•4/30/23, 7:55 PM

after i do blip captions, run the following command - sed -i 's/a man/jrch/g' *.txt and replace man with woman or whatever i want, and jrch with ohwx

ZZono50 Click to see attachment

rafrafOP•4/30/23, 7:57 PM

what are you using here?

Zono50•4/30/23, 7:58 PM

auto1111 with stable-diffusion-webui-ux

Furkan Gözükara SECourses•4/30/23, 7:58 PM

ok here lora vs dreambooth let me know your thoughts . this one lora

Furkan Gözükara SECourses•4/30/23, 7:58 PM

this one dreambooth

Furkan Gözükara SECourses•4/30/23, 7:58 PM

Furkan Gözükara SECourses•4/30/23, 7:58 PM

the training dataset was all taken same day same clothing same bg :d

Furkan Gözükara SECourses•4/30/23, 7:58 PM

i am also doing a new training with included captions

ZZono50 auto1111 with stable-diffusion-webui-ux

rafrafOP•4/30/23, 7:59 PM

oh i see. Havent tried ux yet. reg auto1111 seems ok still. worth using for training in your opinion?

Furkan Gözükara SECourses•4/30/23, 7:59 PM

training dataset

kashy•4/30/23, 8:00 PM

Getting this ERROR after starting training,

" UserWarning: huggingface_hubhuggingface_hub cache-system uses symlinks by default to efficiently store duplicated files but your machine does not support them " .
also says, " Caching files will still work but in a degraded version that might require more space on your disk. This warning can be disabled by setting the HF_HUB_DISABLE_SYMLINKS_WARNINGHF_HUB_DISABLE_SYMLINKS_WARNING environment variable. For more details, see https://huggingface.co/docs/huggingface_hub/how-to-cache#limitations.
To support symlinks on Windows, you either need to activate Developer Mode or to run Python as an administrator. "

and then cache never loads , remains at 0%

Manage `huggingface_hub` cache-system