Software Engineering Courses (SECourses)•17mo ago

@Dr. Furkan Gözükara two models have been changed by the author, %cd /kaggle/temp/VAE !wget "https:/

@Dr. Furkan Gözükara two models have been changed by the author, %cd /kaggle/temp/VAE
!wget "https://huggingface.co/black-forest-labs/FLUX.1-schnell/resolve/main/ae.safetensors" -O "ae.safetensors" !wget "https://huggingface.co/black-forest-labs/FLUX.1-schnell/resolve/main/flux1-schnell.safetensors" -O "flux1-schnell.safetensors" am I right? another question it seems to take a few mins to get 1 image from flux model on kaggle, right?

virginiaOP•8/31/24, 1:35 PM

@Dr. Furkan Gözükara another question, if I want to use custom nodes in swarm UI, I did not see the folder, do I have to create one? Autocompletions Users-log.ldb clip_vision tensorrt
Embeddings Users.ldb comfy-auto-model.yaml unet
Lora VAE controlnet upscale_models
Stable-Diffusion clip diffusion_models, or where are the custom node folder?

m3m3x•8/31/24, 1:38 PM

@Dr. Furkan Gözükara I've test on windows on my PC with A4000 (8go) and it worked, however as I suspected it use shared memory during the step "move text encoder to gpu" as there is not enough memory on the GPU, that's why it OOM on linux, because it can't use shared memory, then it's passed back to the cpu freeing memory and the rest is loaded that require less than 8go so during training it's stay under 8go. Is there a way to not move the text encoder to gpu, like having it calculated on cpu directly, having it use shared memory or making it use less VRAM, so it will work also on linux ?

Vvirginia @Dr. Furkan Gözükara two models have been changed by the author, %cd /kaggle/tem...

Furkan Gözükara SECourses•8/31/24, 1:38 PM

yes links changed

Mm3m3x @Dr. Furkan Gözükara I've test on windows on my PC with A4000 (8go) and it worke...

Furkan Gözükara SECourses•8/31/24, 1:38 PM

i see what you mean now

Furkan Gözükara SECourses•8/31/24, 1:38 PM

so you get error during cache

Furkan Gözükara SECourses•8/31/24, 1:38 PM

you need to ask this to kohya

Furkan Gözükara SECourses•8/31/24, 1:38 PM

cant linux has shared vram?

m3m3x•8/31/24, 1:39 PM

Ok I could maybe create a ticket on it's git

FFurkan Gözükara SECourses cant linux has shared vram?

m3m3x•8/31/24, 1:39 PM

I search but didn't find a way to do that it seem not implemented in linux nvidia driver

Mm3m3x Ok I could maybe create a ticket on it's git

Furkan Gözükara SECourses•8/31/24, 1:39 PM

please make a detailed reply here and tag @kohya-ss : https://github.com/kohya-ss/sd-scripts/pull/1374

GitHub

support SD3 by kohya-ss · Pull Request #1374 · kohya-ss/sd-scripts

Replace SD3Tokenizer with the original CLIP-L/G/T5 tokenizers.
Extend the max token length to 256 for T5XXL.
Refactor caching for latents.
Refactor caching for Text Encoder outputs
Extract arch...

Furkan Gözükara SECourses•8/31/24, 1:39 PM

there are also other expers some of the may chime in

FFurkan Gözükara SECourses there are also other expers some of the may chime in

m3m3x•8/31/24, 1:40 PM

ok thank you I will do that

virginiaOP•8/31/24, 1:43 PM

does it take a few mins in smarmui on kaggle for flux model?

virginiaOP•8/31/24, 1:43 PM

to create 1 image, I mean? seeming very slow

Vvirginia to create 1 image, I mean? seeming very slow

Furkan Gözükara SECourses•8/31/24, 1:44 PM

ye it is slow. kaggle doesnt give good gpus :/

Furkan Gözükara SECourses•8/31/24, 1:44 PM

i wish they started giving bf16

Furkan Gözükara SECourses•8/31/24, 1:44 PM

we could trian flux there

Furkan Gözükara SECourses•8/31/24, 1:44 PM

i trained 24 images on 4x gpu and even 240 epoch is not overtrained. this is impact of effective batch size increase

FFurkan Gözükara SECourses i wish they started giving bf16

Jehex•8/31/24, 1:47 PM

Hi there, may I ask you something, for kohya_ss, for generate the meta_lat.json for the full finetuning of flux, do you know which path of the model should I set as argument, the path to flux1-dev.safetensors seems to not working

JJehex Hi there, may I ask you something, for kohya_ss, for generate the meta_lat.json ...

Furkan Gözükara SECourses•8/31/24, 1:49 PM

i didnt research it yet. hopefully next week

virginiaOP•8/31/24, 1:49 PM

yes I am able to do use swarmUI now, thanks a lot

jhb•8/31/24, 2:03 PM

Whats the best tutorial for Flux image to image generation for product images.?

Jjhb Whats the best tutorial for Flux image to image generation for product images.?

Furkan Gözükara SECourses•8/31/24, 2:04 PM

i never done that but i have shown image to image

Furkan Gözükara SECourses•8/31/24, 2:04 PM

but not for flux

dsienra•8/31/24, 2:10 PM

@Dr. Furkan Gözükara new test with high TE learning rate, now with unique tokens no class and person class regularization, I see that using person class avoid class mixing

dsienra•8/31/24, 2:10 PM

these are my regularizations