i took it and will test
i took it and will test
Fsure
Fgood idea
Fonce you make it work
Fyou can move to l40s
Fyou can even move cached files
Fit shouldnt recache on l40s
GOK I will copy the Lora directory over onced cached
N3 rank
4090 gpu loсal
4090 gpu loсal
Fit shoudl work
Fworking or not?
Noperate
Fnice
Fworking good
Fif you use rank 4 will be faster
Fi am testing torch 2.4.1 right now
Fit may boost even more
Flooks like 2.4.1 still doesnt fix :/
Fok torch 2.5 still best
Ffor windows
Fimproves both rank 3,4,5,6 and all 
Fi am gonna test something else now a new stuff
Frtx 3090 and rtx 3060 faster ranks
Fmin speeds on windows
WIs there a way to change it to use 512x512 images as the source?
Gstill doesnt work, 1 GPU dies in same spot
GI wasted 20$ on multiple instances, I will wait until you test 10,000 IMAGES with Captions and figure out the correct settings
Git does not work
M@Dr. Furkan Gözükara t5 training is out
Mgetting way better results now with captions
M
Fjust change 1024 to 512 on gradio
Fyou should post it as error on kohya meanwhile
Fmaybe it is related to him
Fye i saw
Fbmaltais also merged nice
Fcurrently uses same LR i presume
Myou can't cache text encoder on bmaltais
Fwhat do you mean
MFile "D:\kohya_ss\sd-scripts\flux_train_network.py", line 514, in <module> trainer.train(args) File "D:\kohya_ss\sd-scripts\train_network.py", line 441, in train self.post_process_network(args, accelerator, network, text_encoders, unet) File "D:\kohya_ss\sd-scripts\flux_train_network.py", line 168, in post_process_network raise ValueError( ValueError: T5XXL is trained, so cache_text_encoder_outputs cannot be used / T5XXL学習時はcache_text_encoder_outputsは使用できません
MFi think cache is for when we dont train t5
Fnot sure
Myeah without it, you need 48gb vram gpu
Mgot it working with split mode single
WWhere do I change the 1024 to 512?
Fsearch this keyword on gradio of kohya : Max resolution
WI did that, but it didn't do any training then.
WINFO [Dataset 0] train_util.py:2326
INFO caching latents with caching strategy. train_util.py:984
INFO checking cache validity... train_util.py:994
100%|██████████████████████████████████████████████████████████████████████████████████████████| 14/14 [00:00<?, ?it/s]
INFO caching latents... train_util.py:1038
0%| | 0/14 [00:00<?, ?it/s]18:14:23-407701 INFO Training has ended.
INFO caching latents with caching strategy. train_util.py:984
INFO checking cache validity... train_util.py:994
100%|██████████████████████████████████████████████████████████████████████████████████████████| 14/14 [00:00<?, ?it/s]
INFO caching latents... train_util.py:1038
0%| | 0/14 [00:00<?, ?it/s]18:14:23-407701 INFO Training has ended.
W