Software Engineering Courses (SECourses)•16mo ago

im not sure if its the new cudnns or the torch versions, but i noticed massive quality and learning

im not sure if its the new cudnns or the torch versions, but i noticed massive quality and learning drop off

MMaxivi im not sure if its the new cudnns or the torch versions, but i noticed massive q...

Furkan Gözükara SECourses•9/18/24, 12:09 PM

can totally happen

MaxiviOP•9/18/24, 12:15 PM

i am going to test as well rn

MaxiviOP•9/18/24, 12:54 PM

@Dr. Furkan Gözükara can having different cuda toolkit affect results?

MMaxivi @Dr. Furkan Gözükara can having different cuda toolkit affect results?

Furkan Gözükara SECourses•9/18/24, 1:59 PM

i doubt that

Furkan Gözükara SECourses•9/18/24, 1:59 PM

i seen only different xformers in past

Furkan Gözükara SECourses•9/18/24, 1:59 PM

broken training

Furkan Gözükara SECourses•9/18/24, 1:59 PM

so spda can do same

MaxiviOP•9/18/24, 2:29 PM

you're right i just tested and sdpa gives worse results

Furkan Gözükara SECourses•9/18/24, 2:53 PM

48 gb multi gpu still fails :/

Furkan Gözükara SECourses•9/18/24, 2:53 PM

FLUX: Gradient checkpointing enabled. CPU offload: True
number of trainable parameters: 11901408320

Furkan Gözükara SECourses•9/18/24, 2:53 PM

tried latest accelerator and torch

dxqb•9/18/24, 3:21 PM

why do you offload if you have 48 gb?

Marrvelous•9/18/24, 3:44 PM

Please let me know if this isn't the right place for questions... I followed the tutorial for Swarm UI and to get Flux into Swarm UI. During the installation process I did have an error which resolved when I turned off my Nord VPN connection.

I am able to generate images just fine using Stable Diffusion 3, but when I try either flux model, I receive the following error:

Self-Start ComfyUI-0 on port 7821 failed. Restarting per configuration AutoRestart=true...
[BackendHandler] backend #0 failed to load model with error: System.Net.WebSockets.WebSocketException (0x80004005): The remote party closed the WebSocket connection without completing the close handshake.

Anyone know how to resolve? I've made sure both the Nord VPN app is disconnected and the Nord browser extension as well.

LW24Ai•9/18/24, 5:00 PM

Hello, can you tell me how to reduce the scales of a trained lora flux? At my exit, lora is weighing 1gb

Zono50•9/18/24, 5:07 PM

@Dr. Furkan Gözükara can models do like different font styles? or maybe do like font wildcards for making alogo?

Ddxqb why do you offload if you have 48 gb?

Furkan Gözükara SECourses•9/18/24, 5:11 PM

multi gpu doesnt work even with offload

Furkan Gözükara SECourses•9/18/24, 5:11 PM

multi gpu traning overheard

MMarrvelous Please let me know if this isn't the right place for questions... I followed th...

Furkan Gözükara SECourses•9/18/24, 5:12 PM

hello welcome

Furkan Gözükara SECourses•9/18/24, 5:12 PM

swarmui developer wants you to ask in their discord

Furkan Gözükara SECourses•9/18/24, 5:12 PM

https://discord.gg/smdGbypw

Furkan Gözükara SECourses•9/18/24, 5:12 PM

please ask there

LLW24Ai Hello, can you tell me how to reduce the scales of a trained lora flux? At my ex...

Furkan Gözükara SECourses•9/18/24, 5:12 PM

fp16

Furkan Gözükara SECourses•9/18/24, 5:12 PM

other than that it will reduce quality

Furkan Gözükara SECourses•9/18/24, 5:12 PM

lora network rank 64 32 16

ZZono50 @Dr. Furkan Gözükara can models do like different font styles? or maybe do like ...

Furkan Gözükara SECourses•9/18/24, 5:12 PM

i think flux can do yes

dsienra•9/18/24, 5:19 PM

@Dr. Furkan Gözükara koha training is running much faster with the new version

dsienra•9/18/24, 5:19 PM

there is some news about this?

FFurkan Gözükara SECourses i think flux can do yes

LW24Ai•9/18/24, 5:19 PM

One user wrote to me that my workout has problems due to the weight of 1gb. For some reason, his models are 18 mb each. I have seen such models with a similar weight, but they did not bring me high-quality results and differences