it's really more about steps\epochs than 'how long' since the speed of cards varies so much...
it's really more about steps\epochs than 'how long' since the speed of cards varies so much...
A3000 steps bs 1
D20 epochs 0 resemblance to the training characters using regularization lets see if with more epochs improves.
DTE and unet lr 1e04
JHi, someone know an expert about transformers, diffusers files etc...
UHey! I'm trying torch 2.5 to see if I can speed things up. Think I set it up properly. But I'm not seeing much improvement [some, very little].
25000 steps / 66 hours on a RTX 3090 | 25 images / 5 repeats / 200 epochs
Is that normal for Flux?
25000 steps / 66 hours on a RTX 3090 | 25 images / 5 repeats / 200 epochs
Is that normal for Flux?
Dhow many it/s?
Dwhy are you using 5 repeats? 5 repeats 200 epochs = 1000 epochs is excessive
JThe real photos that I used for training
U9.47
Uhow much'd you suggest? for a style LORA, first test
Dwhat graphics card?
U3090
Dwith T5 attention mask off I get 4.5 it/s in my 3090 ti
Dwith T5 attention mask enabled 6.5 it/s
DI suggest 1 repetition 200 epochs
Uwould you suggest 1 repeat / 200 epochs? 2 repeats / 100 epochs?
Ugreat.
Uthanks!
D5000 steps
Uhave u seen quality differences with / without?
Dno in ma case but is better if you have captions
Dtheoretically
Uso if i'm using captions, i should aswell use t5 attention
Dyes, you can try with and without
Dfor a firts run you can try the fast way 1 repeat 150 or 200 epochs and see the results
Dthen try with attention mask and compare
Djust a suggestion, there is no one solution for every case
Dat this point is trial an error
AAny way to improve these settings for ai toolkit?
message.txt2.23KB
A102 images so far has been going 12 hours
DI only used kohya
Dbut you are using adamw8bit
Dit consumes more vram
DI use adafactor
Dwhat graphic card do you have?
MAnyone use Prodigy for Flux traning?
Dnot me, it too memory hungry
Sare there any differences i should be aware of when making a gguf of a flux.1-dev fp32 safetensor model, compared to the way i would do it for an llm? just looking to make a q8 for people with less vram.
Jusing only original ones from Black forrest labs (Flux1-dev),
there are too many versions out for me to keep track of and whether they are all
compatible with my LoRAs or not.
there are too many versions out for me to keep track of and whether they are all
compatible with my LoRAs or not.
TYeah they are original, you can read the repo page to have the info
N@Dr. Furkan Gözükara 
Congrats on the review, will peruse and need to put likes everywhere
Congrats on the review, will peruse and need to put likes everywhere
Fprobably you loaded into dreambooth tab at least once and corrupted config
Fyes this is a bug
Fyou need to use command
Flike cd checkpoints
Fi didnt try lion
Fyou have to use cfg 1
Fflux has different guidance
