it cant since 1080 doesnt have bf16
it cant since 1080 doesnt have bf16
Fbut 3060 perfectly can
PThank you for letting me know. With 3060 how long does it take to train on average?
Fit would take like 10 hours to get a perfect quality
Fbecause we have to use huge optimization
Fhowever there may come a new technique
Ftraining a single layer
Fwaiting kohya on that
DSeems like vast majority of character Loras work even without including their supposed trigger words. Interesting
Fflux has internal encoding
Fso every image fully captioned
Dnot sure what you two mean. I have used a character lora with flux recently. trigger word (which was the name of the person) was necessary. using the lora without it already did something, but not everything
Fif you dont use trigger word and if it is not overfit you wont get exactly that person
Fbecuase you cant know exact internal encoding / caption 
SDr Furkan i hope evrything is ok, wht do you think could be the best, cost/benefit amount fo dataset image for training?
Fi plan to test with 100 images
Fcurrently works as low as 15 very good
Dhm, does Lora merging work for flux?
Fthey added some tools but didnt test yet
DI remember results weren't always good previously, especially if one wanted to merge loras of different sizes
MWhen using Dreambooth training, what is the formula for total steps when using multiple GPUs? In the video about Flux training it seemed like you would multiply everything by the number of GPUs, so 100 images, 1 repeat, batch size 1, 10 epochs would be 1000 steps on a 1 GPU, 4000 steps on a x4 GPU machine. However, when running a Dreambooth finetune, the system insists it's going to be 1000 steps unless I actually increase the number of epochs to 10 x the number of GPUs…
Fwhatever you calculate divide to number of gpus
Fso single gpu 1000 steps or 100 epochs becomes 250 steps and 25 epochs with 4 gpus
Falso use number of processes qual to number of gpus
MThat's what I thought, but: I had 175 images, so I set 12 epochs hoping I'd get 8400 steps
MBut the command line logs only showed 1200 steps to go
MSo I stopped and changed the settings--should I have let it run with 1200? There's no indication that it's running those 1200 separately on 4 GPU
F1200 steps becomes 4800
Fsince process 4x at once
MSorry, mistyped here: it was supposed to say '2100' of course 
M175x12 → 2100
MSo you're saying that the stdout logs show 2100, but in the end the real count will be 8400?
Sdr Furkan, exist controlnet canny or lineart for flux version that can be used in forgue?
Fyes batch size will be 4
Feffective batch size
Fi think i saw some early ones but probably not good. also i dont know if works with forge
E@Furkan Gözükara SECourses Hello Doctor, I am having problems with the presets tier 1 V2 slow and fast.
I seem to have a memory leak. The VRAM fills up, and then it eats almost all the RAM, the training becomes extremely slow and my PC loses stability so I have to urgently abort the process.
Do you know if Onetrainer updated any parameters? (I just executed the update.bat today) With the other presets or older versions of your own presets don't give any problems. Thanks.
I seem to have a memory leak. The VRAM fills up, and then it eats almost all the RAM, the training becomes extremely slow and my PC loses stability so I have to urgently abort the process.
Do you know if Onetrainer updated any parameters? (I just executed the update.bat today) With the other presets or older versions of your own presets don't give any problems. Thanks.
Dtier 1 is too much for any consumer GPU I think, tier 3 is the best that fits into ANY
MIs there a @Furkan Gözükara SECourses in the house? I couldn't find any material (post or video) about the importance of captioning and regularization images for SDXL Lora/Dreambooth training. Can you point me in the right direction?
MI got very good results with Flux (where you mentioned reg images are not useful, and a single trigger word is good for single persons), but my SDXL results where horrible…
Fyou probably loaded into dreambooth tab
Fand corrupted config
Falso hopefully i will update all configs today
Fit reduces likeliness
Fi tested it for person
FMedium
Check all the images below to see used training dataset and used captionings.
Fye tier 1 is for only 40 GB and above
Eoooh, for 40 GB of VRAM?
Fyep
Fgo tier 3
Frank 3