Software Engineering Courses (SECourses)•13mo ago

<@205854764540362752> can a rank 256 fp16 lora work well if I'm using the fp8 version of my flux dev

@Furkan Gözükara SECourses can a rank 256 fp16 lora work well if I'm using the fp8 version of my flux dev fine tune? or does the lora have to be fp8 as well?

QQM for a multi-gpu system Kohya cannot train. It gives me an error for gpu id not s...

Furkan Gözükara SECourses•12/15/24, 9:22 PM

as far as i know multi gpu jnot working on windows

Furkan Gözükara SECourses•12/15/24, 9:22 PM

but if youy want to train on different gpuıs

Furkan Gözükara SECourses•12/15/24, 9:23 PM

use accelerate gpu id setting

�🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭<@205854764540362752> can a rank 256 fp16 lora work well if I'm using the fp8 ve...

Furkan Gözükara SECourses•12/15/24, 9:23 PM

it should work

ZZhigarev <@205854764540362752> This doesn’t work. I performed LoRA extraction from the m...

Testitou•12/15/24, 9:45 PM

Black screen for me also after conversion to lora fp8

QQM for a multi-gpu system Kohya cannot train. It gives me an error for gpu id not s...

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭OP•12/15/24, 9:52 PM

in Acceleration you can set GPU, and multi GPU, but for fine tune using multi GPU you need 80 GB GPUs, if I remember correctly. No need to edit the bat file

FFurkan Gözükara SECourses it should work

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭OP•12/15/24, 9:52 PM

but it's better to convert it to fp8?

�🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭in Acceleration you can set GPU, and multi GPU, but for fine tune using multi GP...

Furkan Gözükara SECourses•12/15/24, 10:34 PM

true

�🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭but it's better to convert it to fp8? 🙂

Furkan Gözükara SECourses•12/15/24, 10:34 PM

not sure. i keep loras as fp32 and base models fp16

Furkan Gözükara SECourses•12/15/24, 10:35 PM

swarmui is able to auto cast into used precision

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭OP•12/16/24, 7:52 AM

you must have a lot of diskspace

franc aleu•12/16/24, 9:45 AM

Hi Doctor ! @Furkan Gözükara SECourses Please.. do you have some tutorial about training sd3.5? I prefer it to flux for some kind of generations.. I feel it more creative when subjects are not real humans ....

franc aleu•12/16/24, 9:45 AM

thank you

franc aleu•12/16/24, 10:31 AM

as a separate issue, my swarm generates another folder with the name stableswarm so that the model installer for windows does not work well because it generates another parallel swarm folder with another new folder of models ... I think that maybe something is not quite right or does not fit with my folder configuration.

Ffranc aleu Hi Doctor ! <@205854764540362752> Please.. do you have some tutorial about trai...

Furkan Gözükara SECourses•12/16/24, 1:55 PM

i plan to do but i couldnt find time for it yet

�🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭you must have a lot of diskspace 😄

Furkan Gözükara SECourses•12/16/24, 1:55 PM

yep

Ffranc aleu as a separate issue, my swarm generates another folder with the name stableswarm...

Furkan Gözükara SECourses•12/16/24, 1:55 PM

i am not sure but developer replying

Furkan Gözükara SECourses•12/16/24, 1:55 PM

https://discord.gg/hM6CeZmY

zaiar•12/16/24, 2:42 PM

hey quick question, im trying to train a LoRa on Flux, I have a 3060 with 12gb so im using your config tier 2 for and following the youtube tutorial. I disabled all my start ups and my task manager is showing im using about 3-400mb of memory on my gpu when I start training. Ive reduced my resolution to 512x512, increased block swaps, and tried using the 10gb config. everytime I get a CUDA out of memory error and the training fails

zaiar•12/16/24, 2:42 PM

any advice?

ArtChicken•12/16/24, 3:47 PM

does anybody knows if the free kaggle notebook is still good (I'd rather say:"working") at SDXL training? And, if so, a tutorial to use it with the new Kohya interface?

Timson•12/16/24, 5:28 PM

Guys, are the current commit of sd3-flux.1 branch kohya_ss working for you? I swear the branch worked perfectly yesterday, but I'm having this annoying error.

Timson•12/16/24, 5:29 PM

did 2 clean installs with pip cache purging to no avail

Timson•12/16/24, 5:31 PM

also this warning, not shure if they are new or normal

Timson•12/16/24, 5:39 PM

Never mind, I'm just an idiot and forgot --noverify flag, so torch and xformers version reset on gui.sh run. Here goes an hour of my life

D4UNTLESS•12/16/24, 6:57 PM

D4UNTLESS•12/16/24, 6:58 PM

@Timson hi I also used the sd3-flux.1 but am getting this error. You probably fixed this but I don't understand your solution. Could you explain? I'm doing dreambooth training following Furkan's tutorial. I can do lora but on model training, I get this error

DD4UNTLESS Click to see attachment

Timson•12/16/24, 7:30 PM

this is different error, it says CUDA is out of memory, so it probably failed to fit in VRAM whatever you are trying to train

Timson•12/16/24, 7:31 PM

you can use nvitop tool to monitor your GPU status, ensure you have enough of free VRAM to fit all models (flux + t5xxl + clip_l + vae)

Timson•12/16/24, 7:32 PM

around 29 GB for fp16 flux dev

TTimson you can use nvitop tool to monitor your GPU status, ensure you have enough of fr...

D4UNTLESS•12/16/24, 8:03 PM

I used the config that uses 8400mb of vram. I have 12gb 4070 super that had around 10gb free vram. I checked that using nvitop

D4UNTLESS•12/16/24, 8:27 PM

Also, I'm thinking about doing it on massed compute instead. In the tutorial, he said that kohya is not on main branch there. Can I use/switch to this sd3-flux.1 on massed compute?

DD4UNTLESS Also, I'm thinking about doing it on massed compute instead. In the tutorial, he...

Timson•12/16/24, 9:37 PM

I would say doing training in the cloud is the best option unless you can't afford spending a few bucks on this project. It requires more effort though and basic linux administration skills

AArtChicken does anybody knows if the free kaggle notebook is still good (I'd rather say:"wo...

Furkan Gözükara SECourses•12/17/24, 12:13 AM

it works but i cant say sdxl very good

TTimson Never mind, I'm just an idiot and forgot --noverify flag, so torch and xformers ...

Furkan Gözükara SECourses•12/17/24, 12:14 AM

DD4UNTLESS I used the config that uses 8400mb of vram. I have 12gb 4070 super that had arou...

Furkan Gözükara SECourses•12/17/24, 12:14 AM

hello

Furkan Gözükara SECourses•12/17/24, 12:14 AM

how much RAM you have?

Furkan Gözükara SECourses•12/17/24, 12:14 AM

you should upgrade 64 GB and it will fix your issue 100%