Software Engineering Courses (SECourses)•16mo ago

Dr Gözükara, can you shed light on choices of 'Samplers" and their respective scheduler? I usually g

Dr Gözükara, can you shed light on choices of 'Samplers" and their respective scheduler? I usually generate Asians and I have found (not necessarily true) DPM-2 looks better than Euler. Care explain how Sampling works ? (like if you already have 1 in your video)

Jackie•9/4/24, 4:23 AM

I have a question about training LoRA. My LoRA files usually have a very large size after training, while the LoRA files I download from the internet are much smaller. How can I reduce the size of my LoRA files after training?

JJackie I have a question about training LoRA. My LoRA files usually have a very large s...

nicearoni•9/4/24, 4:49 AM

you might try this tab in kohya

FFurkan Gözükara SECourses i will add all later. use this link : https://www.patreon.com/posts/110879657

GomezBro•9/4/24, 5:31 AM

Bro I followed the video, I see at 4 x GPU you do not change your batch size to 4, you leave it at 1. Also, when I hit training, (I have 10,000 images) Changed it to 8 epoch, save each 1 Epoch, I do see that when it starts 2500 steps instead of 10,000 But when I do nvitop I only see 1 GPU being used. 3 are unused or not moving? Why? I thought I would see all 4 GPUs being used. Changed the learning rate to 0.0001 as well on both places.

GomezBro•9/4/24, 5:39 AM

I only see 1 GPU being used

GomezBro•9/4/24, 6:17 AM

(Loaded your 4xGPU-FAST) I see only 1 batch running at 10000 steps now, not 2500. but I do see all 4 GPU being used now in NviTop, GPU 0 fluctuates between 25% and 100% training ends quickly and no model comes out. Its not working at all loading your config. Your config has 0.00005 Learning rate when it loads not 0.0001 like your video shows

GomezBro•9/4/24, 6:17 AM

:1222pepeblunt:

GomezBro•9/4/24, 6:35 AM

Terminated instance so its not charging, will relaunch after you respond bro, thanks

GGomezBro Bro I followed the video, I see at 4 x GPU you do not change your batch size to ...

�

𝔸𝖨ΞӾ•9/4/24, 7:17 AM

Aren't you supposed to list all the other GPU Ids in the command line args? I haven't checked yet but should be already configured in a Multi GPU config if he has one on Patreon

�𝔸𝖨ΞӾ Aren't you supposed to list all the other GPU Ids in the command line args? I ha...

GomezBro•9/4/24, 7:18 AM

his config has that already, i checked it, but his learning rate doesnt match what he does in the video

GomezBro•9/4/24, 7:18 AM

and batch is at 1 not 4 too

GGomezBro and batch is at 1 not 4 too

�

𝔸𝖨ΞӾ•9/4/24, 7:18 AM

Oh ok gotcha, I'm barely watching the video rn

GomezBro•9/4/24, 7:19 AM

theres just a couple conflicts from video/4xgpu gpu configs that you load

GomezBro•9/4/24, 7:19 AM

is a bit grey

GGomezBro theres just a couple conflicts from video/4xgpu gpu configs that you load

�

𝔸𝖨ΞӾ•9/4/24, 7:26 AM

This is from the video, you sure this was configured correctly?

�𝔸𝖨ΞӾ This is from the video, you sure this was configured correctly?

GomezBro•9/4/24, 7:55 AM

thats under accelerated right?

�𝔸𝖨ΞӾ This is from the video, you sure this was configured correctly?

GomezBro•9/4/24, 7:56 AM

I checked it against his video and loaded his settings bro

GomezBro•9/4/24, 7:56 AM

those 4X gpu settings are set

GomezBro•9/4/24, 7:56 AM

that had to be set and was

GGomezBro thats under accelerated right?

�

𝔸𝖨ΞӾ•9/4/24, 8:39 AM

Yeah from Accelerate, what about this setting?

�𝔸𝖨ΞӾ Yeah from Accelerate, what about this setting?

GomezBro•9/4/24, 8:40 AM

Yes all good, i loaded his settings bro

GGomezBro Yes all good, i loaded his settings bro

�

𝔸𝖨ΞӾ•9/4/24, 8:41 AM

Yeah I'm looking at the config template you used, don't know what else to tell ya. I guess wait for the Dr.

GomezBro•9/4/24, 8:41 AM

i relaunched a new instance starting from scratch

�

𝔸𝖨ΞӾ•9/4/24, 8:41 AM

How much faster is training with 4x GPU?

�

𝔸𝖨ΞӾ•9/4/24, 8:41 AM

I'm guessing 4x faster lol

GomezBro•9/4/24, 8:42 AM

his config has 0.00005 Learning rate when it loads not 0.0001 like the video shows too, so its not the same

�𝔸𝖨ΞӾ I'm guessing 4x faster lol

GomezBro•9/4/24, 8:42 AM

yup

GomezBro•9/4/24, 8:42 AM

10,000 Images I think in 6 hrs is what im thinking

�

𝔸𝖨ΞӾ•9/4/24, 8:43 AM

Oh wow damn yeah with dataset that huge it makes sense to go fast

GGomezBro his config has 0.00005 Learning rate when it loads not 0.0001 like the video sho...

�

𝔸𝖨ΞӾ•9/4/24, 8:47 AM

Well he explained why he changed it in the video here

�𝔸𝖨ΞӾ Well he explained why he changed it in the video here

GomezBro•9/4/24, 8:47 AM

Yes but his loaded settings are different

GomezBro•9/4/24, 8:47 AM

and those are set

�

𝔸𝖨ΞӾ•9/4/24, 8:47 AM

but not sure why it's not detecting all your GPUs

GomezBro•9/4/24, 8:50 AM

did all fresh same thing, only 1 GPU in use

GomezBro•9/4/24, 9:00 AM

I get an error after a few mins and it errors out, Training has ended

�

𝔸𝖨ΞӾ•9/4/24, 9:11 AM

Damn bro sorry, Dr should have the antidote ready for you soon

GomezBro•9/4/24, 9:12 AM

there is something off and i cant put my finger on it, because his configs alone should run from loading....

GGomezBro there is something off and i cant put my finger on it, because his configs alone...

�

𝔸𝖨ΞӾ•9/4/24, 9:13 AM

You've successfully trained several Loras with Kohya already, right? I don't remember

�𝔸𝖨ΞӾ You've successfully trained several Loras with Kohya already, right? I don't rem...

GomezBro•9/4/24, 9:14 AM

Yeah on my machines at home i have no issues, perfect setups

GomezBro•9/4/24, 9:14 AM

I have A6000 at home

GomezBro•9/4/24, 9:14 AM

and 3 3090 Ti 24 GB

�

𝔸𝖨ΞӾ•9/4/24, 9:14 AM

Daaayyyuuum

GomezBro•9/4/24, 9:14 AM

I did 1000 Images Lora and its sick, but I want to do 10000 on 4 GPU

GomezBro•9/4/24, 9:15 AM

so it doesnt take long, it should do 10,000 steps in 6hrs

GomezBro•9/4/24, 9:15 AM

and each can be 1 epoch

�

𝔸𝖨ΞӾ•9/4/24, 9:16 AM

Ok I'm communicating with a certified boss playa then lmao so the issue is isolated to the cloud GPU setup then hmm

�𝔸𝖨ΞӾ Ok I'm communicating with a certified boss playa then lmao so the issue is isola...

GomezBro•9/4/24, 9:16 AM

yeah, the config, something is off and crashing on the 4 GPU config or some other issue

МИШАНЯ•9/4/24, 9:45 AM

Greetings to all AI and SD specialists.
I came across the following question. In SD is it realistic to make ONE image from different positions ? For example: Girl: Front, back, side view. At the same proportions, appearance that would not change, I want to create a 2d game, for this I need the same sprites from different positions

Neuromasters•9/4/24, 10:33 AM

And why do we need so many steps, a range of 2k to 4k is not enough for styling on items?

Neuromasters•9/4/24, 10:37 AM

we have a token to bake owix, it is empty, we have described the interactions with it. We gave a prototype that it is an icon object, etc.
steps for retraining other objects in the model memory are missing. Then why such a huge number of steps?

Ffoobax Any tips on getting more realistic results? You've replicated the face very well...

Furkan Gözükara SECourses•9/4/24, 12:08 PM

use more and better training images

Dr Gözükara, can you shed light on choices of 'Samplers" and their respective scheduler? I usually g

Similar Threads