L40 is slightly faster
L40 is slightly faster
AHmm, it also has more storage, so maybe I should switch to a 1xL40 server then?
I have 53 hours left with only 1 hour passed. Think I'll reduce the time (and it's also slightly cheaper than 2xA6000) by switching?
I have 53 hours left with only 1 hour passed. Think I'll reduce the time (and it's also slightly cheaper than 2xA6000) by switching?
Fyes i recommend
AWill do
AAlso, have you considered perhaps creating a Docker image with all the tools already included, as it seems that Massed Compute (and I assume other services like Runpod as well) support Docker images?
It'll both simplify the process (no need to download stuff and install, and users less likely to encounter an error if for example a Python package is updated), and save time (which cost money when renting).
I don't know however if running through a Docker container affects the performance when using a GPU
It'll both simplify the process (no need to download stuff and install, and users less likely to encounter an error if for example a Python package is updated), and save time (which cost money when renting).
I don't know however if running through a Docker container affects the performance when using a GPU
Fwe already have massed compute image and it is already big
Fall scripts would be just huge
Fbest to install them because installation is super fast on massed compute
PI was able to make it work just need to read more the documentation and the warings for a local 2x3090, no nvlink
Pnot shure but when using multi gpu it seens to be slower than using just one, not shure if is a bad config or a comunication overhead
Fremember you will do half epoch
Fwhen doing 2x gpu
Fit will be same
Fbecause your batch size becomes 2
PDoes the global amount of steps change?
Pthe batc/epoch is smaller but the overal amount is slower
DHi Furkan, are you referring to a flux finetune in Koyha, and not a dreambooth training?
If so, what is the main difference between dreambooth and fine tuning? Fine tuning is more for style/concept version dreambooth being better for specific subjects? Thanks
If so, what is the main difference between dreambooth and fine tuning? Fine tuning is more for style/concept version dreambooth being better for specific subjects? Thanks
BI made a new tutorial for my Everly Heights ClipFrame tool. Video editors, producers, and people building datasets from videos will find it useful.
https://youtu.be/qrv9vgd1WH0?feature=shared
https://youtu.be/qrv9vgd1WH0?feature=shared
YouTubeBill Meeks
Discover Everly Heights ClipFrame, a versatile tool designed to revolutionize how professionals and creators analyze, organize, and extract value from video content. Perfect for marketing teams, content creators, video editors, and even AI developers, ClipFrame simplifies the process of breaking down videos, organizing frames, and exporting cust...
Fit doesnt keep previous training logs
Ftherefore you have to reduce new epoch
Fso 200-170 if you are using 170 epoch base
Fno different atm
Fdreambooth difference is using regularazation images
PIt's safe to say that gor low epoch count Muti gpu has a little impact ?
Whi @Furkan Gözükara SECourses do you have plan to train Flux Schnell model ? I heard they have some great model. For example this model: https://civitai.com/models/943001/shuttle-3-diffusion
DHi Furkan, I've been experimenting with Flux dreambooth and LoRA training for several months now, using a high quality dataset of 3D renders en photographs of a specific car (which is not trained in the base model).
After extensive testing, I am actually getting better results with Flux LoRAs then dreambooth finetune.
The finetuned model does not follow the car's body well, the LoRA is replicating the car's shape a lot better.
For the LoRA, I found that AdamW with network dimension 128 and alpha 64 worked best for me.
For the dreambooth, I've used your configs and left them untouched.
I am surprised that this is the case as you would expect the oposite. Even the extracted LoRAs from the trained checkpoint performed worse than the LoRA.
I know this is very little information to go on, I can send you more grid tests if you want, but I am wondering if you have an idea where to look.
The only thing I can think of is that the LoRA uses AdamW and dreambooth uses Adafactor.
I could not use AdamW for dreambooth, not even on a 48GB GPU.
My goal and quality metrics are product (car) accuracy. So perhaps my configs should differ from your configs that are more leaned towards people?
PS. I've left all the common settings the same like image repeats, dataset/captions, noise offset, snr. And I have tested multiple epochs to find the best sample size
After extensive testing, I am actually getting better results with Flux LoRAs then dreambooth finetune.
The finetuned model does not follow the car's body well, the LoRA is replicating the car's shape a lot better.
For the LoRA, I found that AdamW with network dimension 128 and alpha 64 worked best for me.
For the dreambooth, I've used your configs and left them untouched.
I am surprised that this is the case as you would expect the oposite. Even the extracted LoRAs from the trained checkpoint performed worse than the LoRA.
I know this is very little information to go on, I can send you more grid tests if you want, but I am wondering if you have an idea where to look.
The only thing I can think of is that the LoRA uses AdamW and dreambooth uses Adafactor.
I could not use AdamW for dreambooth, not even on a 48GB GPU.
My goal and quality metrics are product (car) accuracy. So perhaps my configs should differ from your configs that are more leaned towards people?
PS. I've left all the common settings the same like image repeats, dataset/captions, noise offset, snr. And I have tested multiple epochs to find the best sample size
RI would be happy to test your LoRA configuration on a character/person if you're willing to share it. I have fine-tuned models and created LoRAs for a couple of months and my findings match Furkan's results, so it would be nice to see if there is a better configuration for LoRAs.
DI'm happy to share my json or toml file in a message (maybe dropping it here is against the community guidelines?).
Keep in mind that I am absolute not a ML expert, I am a 3D product visualization artist dabbling with AI. However I have a beefy pc (4090 + 3090 + 3080) so I've been able to do a lot of testing, running 2 trainings at the same time.
Most of my "knowledge" comes from using custom Claude & Perplexity project with documentation about LoRA, koyha etc.
In any case, I am also very surprised by these results, as I am sure Furkan has found the best configs
Keep in mind that I am absolute not a ML expert, I am a 3D product visualization artist dabbling with AI. However I have a beefy pc (4090 + 3090 + 3080) so I've been able to do a lot of testing, running 2 trainings at the same time.
Most of my "knowledge" comes from using custom Claude & Perplexity project with documentation about LoRA, koyha etc.
In any case, I am also very surprised by these results, as I am sure Furkan has found the best configs
Feloborate more
Fi tried and it didnt work well
Fi dont see any reason to use it for individuals
Finteresting
Fi never had such case
Ffine tune always yields better and it is also expected
DCould it have something to do with the Adafactor versus AdamW optimizer?
P@Furkan Gözükara SECourses like the fist one is as single gpu and will take 150 hour the second one in dual gpu is expecte to take 250 hpurs
Fadamw is better than adafactor at quality slightly. but it uses more vram
Fwhen you do 2x gpu
Fyou need to reduce number of epochs 1/2
Fif 3 gpus
F1/3
Pok thanks, I think I understand it better now
Fyep because multi gpu increases batch size but keep same step count
Fso 1 step you actually do 2
DMaybe that was the difference then as I trained the LoRAs with AdamW but the dreambooth finetune with Adafactor (I could not fit AdamW in my gpu or a rented 48GB one)
Do you think this could be the difference?
I still believe finetuning should be better so I’m just confused about why LoRA yields better results for me.
Do you think this could be the difference?
I still believe finetuning should be better so I’m just confused about why LoRA yields better results for me.
Fhard to know possibly a mistake somewhere 
AHey there!
Anyone out here was deploying flux lora training as an API?
Would be happy to chat or be grateful if you can share something useful on how to deploy to runpod/other compute providers.
P.S. If you can share some useful presets for training person lora - beer on me haha!
Anyone out here was deploying flux lora training as an API?
Would be happy to chat or be grateful if you can share something useful on how to deploy to runpod/other compute providers.
P.S. If you can share some useful presets for training person lora - beer on me haha!
HDoes training lora for Noobai v-pred require different settings?
I'm using kohya_ss gui (dev branch that is up to date with the latest script).
Someone somewhere said that it needs
I'm using kohya_ss gui (dev branch that is up to date with the latest script).
Someone somewhere said that it needs
--zero_terminal_snr
Fwhat is Noobai v-pred
Wrecently when people train lora with kohya and they also train clip text encode together, they claim that lora train together with clip text encode give much better result. I wonder your latest lora training script json file have these clip text encode trained with lora or not ?