download v3 zip file please
download v3 zip file please
Fand
Fuse Windows_Install.bat
FFdelete wan v2 folder
F
Patreon
Get more from SECourses: FLUX, Tutorials, Guides, Resources, Training, Scripts on Patreon
Ftorch vision arrived amazing
Fnow all i have to do is compile flash attention
JHi guys Im new here so please let me know if this question doesn't go here. But Im trying to train lora's using fluxgym I have a couple of questions / issues -
- My training datatset is of kids (age 5-13) with max 15-20 photos. But even with an A100 40GB it takes around 2hrs to train. Any fix for that? I love how fal.ai does the fast training Im okay with even an hour but 2-3 gets tiresome
- I tried with increasing batch size for perfomance but seems like they kinda ruin the quality. Any advice to how to maintain quality with perfomance? captions yay or nay? size 1024 or 512?
1does that mean you can stretch the 5090's legs soon. I haven't been able to test the app yet with work getting in the way.
Fit is quality vs time
Fi am trying to compile flash attention noıw
1ohh wow, things are starting to move along nicely
Fyep
F
Fflash attention compile lets see i hope works
CFP8 will be extra slow on any gpu before the nvidia Lovelace generation
like A6000, 3090, A100 dont have the chips needed to perform well with FP8
L40, 4090, h100 would be good cards for FP8
just incase anyone cares
like A6000, 3090, A100 dont have the chips needed to perform well with FP8
L40, 4090, h100 would be good cards for FP8
just incase anyone cares
JAny idea how fal does it? the speed is amazing & quality is almost perfect too. Are they finetuning layers or just good GPUs
Fnice info thanks
Fyes i know i made a research for a client
Fhigher batch size bigger GPU multiple GPU
Flower steps higher LR
F
Tincrease learning rate until you are no longer satisfied with quality
Tlowering resolution is also an option
Talso decreasing dataset diversity will speed up likeness gain at the cost of generalization
Z14b i2v on default settings with 24gb Vram preset is almost 1hr30 to generate the video, oooof
F
Fit takes 3 hours on 3090 ti
Fi hope this works time to test 5090
Z1.3b txt2video is actually far faster than i would have thought
Zunder 2 mins
Fyep super fast
Fand super high quality
Fyes flash attention compiled
Fnow gonna test
TIs it possible to create lora in wan?
Fnot atm i think
Frtx 5090 speed native python 3.10 venv
Fawesome 5090 workng
Famazing
Fnow i can update all of our apps
1Thats not terrible..
Flet me test fp8 if will work
Fthis is bf16
F522 watt pretty good utilization
FRTX 5090 vs 3090 TI
1Nice..
Fyep
