Software Engineering Courses (SECourses)•3y ago

that means you skipped this step

Alamin H•5/1/23, 7:03 PM

and what about these cells

Furkan Gözükara SECoursesOP•5/1/23, 7:03 PM

please restart begining and execute every cell with order

Furkan Gözükara SECoursesOP•5/1/23, 7:03 PM

if you skip a cell it wont work

FFurkan Gözükara SECourses that means you skipped this step

Alamin H•5/1/23, 7:03 PM

i did that

FFurkan Gözükara SECourses that means you skipped this step

Alamin H•5/1/23, 7:03 PM

Furkan Gözükara SECoursesOP•5/1/23, 7:04 PM

in your code ther eis no gc

FFurkan Gözükara SECourses if you skip a cell it wont work

Alamin H•5/1/23, 7:04 PM

but i dont know why i see some new cells that i did not see on video

Furkan Gözükara SECoursesOP•5/1/23, 7:05 PM

i will update link wait a sec

Alamin H•5/1/23, 7:05 PM

okay sir :4827spidermanok:

AAlamin H okay sir :4827spidermanok:

Furkan Gözükara SECoursesOP•5/1/23, 7:09 PM

thanks for bringing into my attention here

Furkan Gözükara SECoursesOP•5/1/23, 7:09 PM

https://www.kaggle.com/furkangozukara/deepfloyd-if-4-3b-generator-of-pictures-video-vers

DeepFloyd-IF-4.3B-Generator-of-Pictures Video Vers

Explore and run machine learning code with Kaggle Notebooks | Using data from T5 prompts DeepFloyd IF

Alamin H•5/1/23, 7:32 PM

yeah its working perfect now

Alamin H•5/1/23, 7:32 PM

mayen•5/1/23, 7:38 PM

@Furkan Gözükara SECourses can you help me solve this, ive been at it for 2 days i still havent been able to use kohya

mayen•5/1/23, 7:38 PM

mayen•5/1/23, 7:39 PM

RuntimeError: CUDA out of memory. Tried to allocate 20.00 MiB (GPU 0; 8.00 GiB total capacity; 7.20 GiB already allocated; 0 bytes free; 7.29 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation.

Mmayen Click to see attachment

Furkan Gözükara SECoursesOP•5/1/23, 7:47 PM

you need to enable memory optimizations

Furkan Gözükara SECoursesOP•5/1/23, 7:48 PM

looks like you already did

Furkan Gözükara SECoursesOP•5/1/23, 7:48 PM

what is your batch size?

Furkan Gözükara SECoursesOP•5/1/23, 7:48 PM

it is also 1

Furkan Gözükara SECoursesOP•5/1/23, 7:48 PM

i see

Furkan Gözükara SECoursesOP•5/1/23, 7:49 PM

which torch version did you chose @mayen

Haashim•5/1/23, 7:52 PM

Very cool deepfloyd video

Haashim•5/1/23, 7:52 PM

Also I believe I can try your Lora video with colab @Furkan Gözükara SECourses

Haashim•5/1/23, 7:52 PM

That notebook is a bit tricky to me

HHaashim That notebook is a bit tricky to me

Furkan Gözükara SECoursesOP•5/1/23, 7:54 PM

agreed

Furkan Gözükara SECoursesOP•5/1/23, 7:54 PM

actually i prefer gradio over these but they dont release and i dont get why

Furkan Gözükara SECoursesOP•5/1/23, 7:55 PM

i even written them from twitter multiple times

FFurkan Gözükara SECourses which torch version did you chose @mayen

mayen•5/1/23, 7:59 PM

i chose the one that was NOT 2.0, my batchzie was small wither 2 or 1 eveytime i ran it, 512 x 512 images

Mmayen i chose the one that was NOT 2.0, my batchzie was small wither 2 or 1 eveytime i...

Furkan Gözükara SECoursesOP•5/1/23, 7:59 PM

could you have anything else that is consuming your GPU vram?

Furkan Gözükara SECoursesOP•5/1/23, 7:59 PM

what is your GPU model

FFurkan Gözükara SECourses what is your GPU model

mayen•5/1/23, 8:00 PM

rtx 3070ti 8gb ram, no programs running

mayen•5/1/23, 8:00 PM

Furkan Gözükara SECoursesOP•5/1/23, 8:00 PM

can you try with clip skip 1? i dont know if that would make difference.

FFurkan Gözükara SECourses can you try with clip skip 1? i dont know if that would make difference.

mayen•5/1/23, 8:01 PM

whats a clip skip exactly?

Furkan Gözükara SECoursesOP•5/1/23, 8:02 PM

it is about training weight vectors

Furkan Gözükara SECoursesOP•5/1/23, 8:02 PM

you made it 2

Mmayen whats a clip skip exactly?

Furkan Gözükara SECoursesOP•5/1/23, 8:03 PM

could you show screenshots of the all settings if possible

FFurkan Gözükara SECourses could you show screenshots of the all settings if possible

mayen•5/1/23, 8:03 PM

yes

mayen•5/1/23, 8:04 PM

Mmayen yes

Furkan Gözükara SECoursesOP•5/1/23, 8:07 PM

make clip skip 1 and check these 2

Furkan Gözükara SECoursesOP•5/1/23, 8:07 PM

Furkan Gözükara SECoursesOP•5/1/23, 8:07 PM

also are you using standard lora? what network size?

FFurkan Gözükara SECourses make clip skip 1 and check these 2

mayen•5/1/23, 8:12 PM

same problem

mayen•5/1/23, 8:13 PM

TheRustingDeadAdmin•5/1/23, 8:14 PM

still having same issue where regularization images come out looking like sketches and training ends up looking artifacted no matter how long/short the training session

Mmayen same problem

Furkan Gözükara SECoursesOP•5/1/23, 8:14 PM

could it be trying to use your second gpu? amd?

TTheRustingDeadAdmin still having same issue where regularization images come out looking like sketch...

Furkan Gözükara SECoursesOP•5/1/23, 8:14 PM

with all models?

Furkan Gözükara SECoursesOP•5/1/23, 8:14 PM

or certain model

TheRustingDeadAdmin•5/1/23, 8:14 PM

followed your videos to the tee, any model

that means you skipped this step

Similar Threads