can we use our lora with gguf quantized models of dev?
can we use our lora with gguf quantized models of dev?
Tthe problem is with --always-gpu @Dr. Furkan Gözükara
Zwhat's the recommended way to run inference with low gpu cards? we can't use flux1.dev
KTrained a lora with AI-toolkit and the outputs look good, but when I try to run it in Forge it doesn't at all look like the results in Ia-toolkit :/
KWhat am I missing
KI have trained on fp16 dev and are usiing dev nf4 in forge. Is that why?
KCan't trained LORA's be run in Forge? -
Fwhich config you using?
Fi havent tried yet
Fbut swarm ui should work out of box with fp8
Fnf4 is the problem
Fmost optimized
Rhey @Dr. Furkan Gözükara using swarmui grid generator, if run a grid gen with two params, and you rerun the same params but add an extra value, does it overwrite the entire previous gen? I was under the impression that simply adding a value in an existing param, just inserts the extra gens into the existing grid
Fi am not sure
Fi think it has mixed behaviour
Rits really comfusing. using on runpod. ive tested on smalll images to see the mechanism, and it seems sometimes erratic, unpredictable. i am almost sure this the case
Rbut one thing i d like to know is, if i interrupt the grid genartion, can i re-run it? wi ll it skip already generated images? Or does the interruptioon corrupt the files? Surely you've tested and experienced the behavior?
Fit can resume
Z@Dr. Furkan Gözükara in kohya can we puase and resume traininngs?
Fcurrently it doesnt work i think
Fwith fine tuning
Fyou can use last check point but people said
KNoticed
Flora doesnt work
Fi didnt test
Zah ok
Ftime to make some image show :d
F40 images posted here : https://medium.com/@furkangozukara/ultimate-flux-lora-training-tutorial-windows-and-cloud-deployment-abb72f21cbf8
Medium
I have done total 104 different LoRA trainings and compared each one of them to find the very best hyper parameters and the workflow for…
PHow to emphasize keywords that need training in the prompt? Would it be more emphasized if I caplock the keyword or put it in ""?
Fyou mean improve their importance?
Fi think you can write multiple times
Fbut it depends on model
Ffor flux it has internal captioning works amazing
PThat's right, I want to increase the importance of keywords. I'm training on flux to see if capping or leaving it in "" is effective. Or just need each image to have a keyword?
Fgood question sadly i didnt test
Fflux doesnt have a research paper
Fso only way is testing
PI understand, let me try testing with each image having the same keyword
МIs there a flux IP ADAPTER FACE?
I want to transfer an existing (real face) Into a cartoon style...
I want to transfer an existing (real face) Into a cartoon style...
Fi heard someone trying but didnt saw yet
МIf there is information, where can I find it? I really need it.((
Fi dont know atm
Sdr furkan I have a question I used the rank 5 setting on a first model with 200 epoch and the result was stellar I tried another model with the same amount of images, another character and the result was good but not as good as the one before this one I only put 125 epoch can that change in value affect the final result? I ask this because the first one, the one that was real good, the best lora was 75 I thought that if I lowered the epoch it would result in having a good checkpoint before epoch 100 is that sum wrong? is it better to leave the 200 epoch as the rank 5 setting commands?
AWhat are rank meanings in flux training?
Fi think best is go up to 200 and compare check points
Frank is quality
Fof my configs
Sa ok thanks dr Furkan