Oh, L40S will be faster, but quality output will be same as a6000 yes?
Oh, L40S will be faster, but quality output will be same as a6000 yes?
Fbut A6000 is the most price / performance
Fon massed compute
οΏ½Oh, so if I want fastest possible training, it's the h100.
οΏ½rented 1 comp for a month with a6000, at the end of the month, it was so slow, totally unresponsive, i was not even able to do a restart, so had to delete the deployment and redeploy. Maybe I should have restarted the machine more during that month, or maybe you're not supposed to rent it for that long in one run.
οΏ½massed compute
Fthey rarely have bad machines
Fif you report machine id here
Fwe can ping the owners
οΏ½ok thank you, I probably can't find that id now, when it's deleted?
Fyou cant. when that happens report here and ping me :d
οΏ½thank you! btw, should I still use your downloader zip file for kohya when using massed compute? or just run kohya update then run kohya directly from massed compute?
Fbest and most up to date
οΏ½excellent, was just about to start, 91 character training pics. 1 character. 200 epochs save at every 25 should be ok? my last character train with 66 pics, when comparing epocs for realism, 25, 50, 75, 100, 125, the last one, epoch 125 was the best, with 100 being the 2nd best.
F200 could be overfit for 91 images
οΏ½so 150?
F150 good
Fand compare checkpoints
οΏ½yes, save every 25
Fbut if you wna to be 100% sure 200 way to go
οΏ½sure, for research then
οΏ½200, save every 25, then do the amazing prompts and grid you taught us 
thank you for the quick reply, no one works harder than you in AI, and we are grateful!
thank you for the quick reply, no one works harder than you in AI, and we are grateful!Fyou can also save like every 20 or 15 depending on storage
οΏ½how much faster is the h100 compared to the a100 to complete fine tuning?
οΏ½i'm taking the h100 now, can try a100 next time.
Fbut this looks wrong
Fwait
Fok it looks true
FL40s and H100 same
Fso L40s would be faster than a100
Factually best case looks like comparing all 3
οΏ½ha, so more flops means higher speed. how about dual h100, would that make it faster or not?
οΏ½The speed of H100 GPU , 2.54s/it .
Freally good
Fbut dont worth it i think
FA6000 gets 6.3 second
Fso it is like 2.4 times faster
Fbut how many times more expensive
Ftry L40S
οΏ½awesome, I will 
Yeah, I'm overpaying atm, but at least I can use my checkpoint tomorrow noon
Yeah, I'm overpaying atm, but at least I can use my checkpoint tomorrow noon οΏ½L40S, got it, thanks!
X@Furkan GΓΆzΓΌkara SECourses Is it possible to put a new optimizer? like this one?https://github.com/derrian-distro/LoRA_Easy_Training_scripts_Backend/blob/main/custom_scheduler/LoraEasyCustomOptimizer/came.py
I saw few people use this optimizer and another thing called REX
I saw few people use this optimizer and another thing called REX
GitHub
The serverside backend created for use with the LoRA Easy Training Scripts Frontend - derrian-distro/LoRA_Easy_Training_scripts_Backend
Xlike this:- ss_optimizer": "LoraEasyCustomOptimizer.came.CAME
Xcreators who train characters use this, like to know how it different from others
S@Furkan GΓΆzΓΌkara SECourses doc, what's limiting my epochs to only 18 and 1600 steps? using the new 48gb configs in kohya.
Fyou can request from kohya only he can do