how much vram is it using?
how much vram is it using?
Fi am using max in all cases
P@Furkan Gözükara SECourses hello ,I am trying to fine tune SDXL with LoRA with 23 high quality photos of a same person on civitai .Please let me know what parameter should i set.I am new to this so pleae let me know the values I should enter in each box.Thanks
Ftry to match with our kohya config
Pok
Rwhich is currently the best image to video model (free or paid, doesn't matter, just wanna know which is least bad!)
Fthe best ones are paid ones but i dont know which one best
Fkling really good
RI tried luma labs ray 2 and its not good
Rnearly every gen descends into some kinda terrible random motion nonsense, like it might do 2 seconds ok then suddenly someone goes flying through a wall or something stupid
Roccasionally it does something decent but a lot of times it just ignores what you ask and shit flies all over the place randomly, helicopters suddenly turn inside out and backwards lol
Rjust asking it to fly over a building following a helicopter lol
Fdoing new research https://www.reddit.com/r/SECourses/comments/1iqge85/started_doing_new_training_research_on_flux_dev/
Reddit
Explore this post and more from the SECourses community
DI'm trying out massed compute with your tutorial and the thindrives option says it's broken. any idea what that means and how to correct?
.@Furkan Gözükara SECourses I’m currently training Flex on my dataset and I feel it learns much quicker with half the steps compared to dev dedistilled. You were mentioning the learning rate might need adjustment. Do you think it needs to be lowered?
.After only 4000 steps. Not even finished the whole epoch.
A@Furkan Gözükara SECourses Hey so I experimented today a bit and got lost . Can you write what a segment syntax example looks like? Like is it 2 <segment> for -1 and -2?
What would the format look like ?
What would the format look like ?
AFor 2 Lora faces
Fif learning true fast yes
Flower needed
Fyes when connecting
Fselect end existing session
Fbut it terminates all running apps
F
Ai did didn't get an answer yet
Awill it be like <segment:yolov8.pt-1,0.6.05a photo of OHWX man <Lora:Testman-1>>. ?
Fthat seems wrong to me
Fprompt like this
F<segment:yolo-face_yolov9c.pt-1,0.7,0.5>photograph of ohwx man
Fand then click lora
Aand for Lora?
Aoh ok
Fbut actually i dont know how to use lora
Fon prompt 
Fbut it should come at the end
Fappended
Ait say in the document after the text prompt to type manually<Lora...>
A<segment:yolo-face_yolov9c.pt-1,0.7,0.5>photograph of ohwx man <Lora....>
Ayeah I think that's jt thanks doc !
Fye
TDo you meat at training? At inference L40S is only 8% slower but 0.3 second new prompt delay opposed to >3s on 5090
Tso for practical purposes L40S is still better
Fi think it depends if you are constantly changing prompt yes
Fbut L40S is 8000+ USD
Fand 5090 is at its slowest performance
Fit will only get better we dont even have official pytorch for windows
P@Furkan Gözükara SECourses Hello I just started training with the 24 GB flux dev model but my storage was less only 50gb so i think it can create only 2 checkpoints but I checked the training status and the second checkpoint epoch has already strated even before checkpoint one epochs are totally complete.How to save one checkpoint to huggingface if the second one has also started simultaneously
P@Furkan Gözükara SECourses and also once the storage is full what error will I get and how to resume the training
P@Furkan Gözükara SECourses actually I have set 4 checkpoints instead of 2
Fi never could make it save to hugging face before automatically from kohya