does kohya work with rtx 5090? 🤔
does kohya work with rtx 5090? 
Fyes
Fwe have updated our installers just use it
Fjust use your last checkpoint 
Fi never used last state
Fwith our config exactly same
QThanks! i've just upgraded my setups from 4090 to 5090 and noticing that I have problems running pretty much everything I tried before 
thanks again! Now i think im getting it
i've just upgraded my setups from 4090 to 5090 and noticing that I have problems running pretty much everything I tried before thanks again! Now i think im getting it
HHey, I wonder, has anyone tried / have article or any sort of content about full checkpoint finetuning on Wan ? I really want to get my hands into it but it'd be great to have some references to start working on it
Has far as I know only DiffSynth offers the possibility to full checkpoint training on Wan, right ?
Fgood point. i havent checked that yet though. currently musubi is good for lora
�What are the optimal sdxl fine tune and lora configs? Having a hard time finding them. Using massedcompute. Onetrainer or dreambooth? Was thinking of using big love xl 2.5 as base checkpoint. Like the realism and quality of pics there.
�
Fi have dreambooth config
Fbut not for lora. for lora i recommend do dreambooth and extract
�Patreon
Get more from SECourses: FLUX, Tutorials, Guides, Resources, Training, Scripts on Patreon
Fyes this is for kohya
FPatreon
Get more from SECourses: FLUX, Tutorials, Guides, Resources, Training, Scripts on Patreon
MI saw that in 24GB_GPU_Quality_Tier_4, it's configured with "blocks_to_swap": 0,
but in 24GB_GPU_Quality_Tier_1, it's set to "blocks_to_swap": 29.
Theoretically, wouldn’t using 29 blocks to swap reduce quality/consistency?
Or is it necessary to use it, otherwise there would be a VRAM impact?
but in 24GB_GPU_Quality_Tier_1, it's set to "blocks_to_swap": 29.
Theoretically, wouldn’t using 29 blocks to swap reduce quality/consistency?
Or is it necessary to use it, otherwise there would be a VRAM impact?
MI'm just getting started with the more technical parts, so I might be saying something wrong if that's the case, I apologize.
JHi everyone! Great to be here—thanks for all the amazing tutorials and shared resources.
I’m currently running into an issue with my first Flux LoRA training on Koya. I’m on an RTX 3080 (16GB VRAM) and followed the install/setup steps as advised. All required models are downloaded. Everything seems in place, but when I try to run the training, kohya starts up, seems to run ok then without warning goes black, crashes and restarts. It’s happened several times now.
Not sure what I’m missing—or doing wrong? any advice would be hugely appreciated. Thanks in advance!
I’m currently running into an issue with my first Flux LoRA training on Koya. I’m on an RTX 3080 (16GB VRAM) and followed the install/setup steps as advised. All required models are downloaded. Everything seems in place, but when I try to run the training, kohya starts up, seems to run ok then without warning goes black, crashes and restarts. It’s happened several times now.
Not sure what I’m missing—or doing wrong? any advice would be hugely appreciated. Thanks in advance!
Fblock swap only changes vram usage - thus speed not quality
Fhow much RAM you have?
Fthis sounds like out of RAM error
MOh got it, thanks a lot for clearing that up!
Fyou are welcome
MOh, just one more thing. For example, is it possible to 'pause' a LoRA at checkpoint 50 and then resume from that point to continue training up to 100 or even 200?
Fit has use last lora weights
Fon interface you can try
FMAh, got it. I'll give it a try. Thank you so much!
JHello, it’s an eGPU RTX3080 with 16GB RAM, task manager says it’s all available. I noticed my HDD had dropped to 50gbs free space, I’ll free more up and try again.. should I try with one of your 12gb Lora best wf’s instead?
JI always believed that VRAM was the same as RAM? This party always confuses me!
Ftotally different
FRAM is system RAM and VRAM is GPU RAM
Fyou can show me screenshot of task manager
4Vram is a type of ram only for the gpu, it has way better bandwitch and optimized for certain type of data and lot of transfer. But it has a lot more latency that the ram and it too slow for random acces file and slow data that's why we got ram and vram
4DDR5 in dual channel can transfer like 90GB/s, the GDDr7 of the 5090 can do 1792GB/s. Nothing compare
4@Furkan Gözükara SECourses
Hello,
i run a dreambooth training on my machine and when i leave my home and come back i saw the training crash with no explication. I have a 10epoch checkpoint, (i think i was near 15), how i can resume the training ?
Hello,
i run a dreambooth training on my machine and when i leave my home and come back i saw the training crash with no explication. I have a 10epoch checkpoint, (i think i was near 15), how i can resume the training ?
JHey thanks for sharing the info on Vram , I feel that I understand it now but still not entirely sure how to measure what Vram I have! I’ve attached a screenshot from my task manager maybe someone can take a look, knowing what Vram I have will help me choose the correct Koyha Flux workflow, thanks in advance!
Fprobably out of RAM
Fyes you can give last saved checkpoint
Fand it will continue - just reduce number of epochs you gonna do
Fyes you have 32 GB RAM that is the issue
Fplease upgrade to min 64 GB
Fmore is better
4You have 16GB vram, it's write in "Dedicated GPU memory"
JThanks for checking that for me! I’ll probably take the good Doc’s advice and upgrade soon — but in the meantime, out of the Lora training workflows available, which one would you recommend I try with my current setup?
Fsadly none will work very possibly but you can try this
Fuse fp8 dev flux
