it's a kaggle problem
it's a kaggle problem
Fi updted kaggle notebook
Fnow works better
Fwe use working directory
Ftemp is too slow
Ffails
Athe same for the patreon?
Fyes on patreon
Fare you using latest version?
SI am using sd-scripts directly for training and then using diffusers for inference.
I’ve narrowed down the issue to bf16 mixed precision setting not loading the vae as half precision.
I’ve narrowed down the issue to bf16 mixed precision setting not loading the vae as half precision.
SNot clear to me why the vae loading doesn’t support bf16 and instead loads the full float version
Fdiffusers inference problematic
Fi am giving consultation to 1 company
Fthey failed to replicate automatic1111 yet
SPossibly, but seems like the issue is with the training at the moment
SWas going to change the mixed precision to use fp16, but it’s incompatible with full_bf16 training.
SSo I’m either missing something or the vae for bf16 training requires full precision
Fvae is fp16 or fp32
Fbf16 vae probably wont work
Fnot engough precision
Falso do you have bf16 vae?
SI don’t. I’m just using the one from hugging face that you linked to in a tutorial.
Fthat is fp16
Fnot bf16
SI’ll have to look at your dreambooth settings again (on my phone), are you using full bf16 training?
SAnd mixed precision bf16?
SOr just saving the model as bf16?
Fdreambooth training is full bf16
Fbut that is indepented of vae when we do inference
Fyou can also use embedded vae
Fwe dont train vae
Sah ok, so you don't set a different vae when training? You just use the one embedded in the model?
Sthat is probably where my confusion is coming from -- i'm setting the one that you specified for inference for training also
Fno i never set vae when training
THello! Which extension do you recommend for face swapping? Only image, no video
Fi only used roop but not as an extension
SHi @Dr. Furkan Gözükara I have the following message on kaggle and I cannot type "y" N: This must be accepted explicitly before updates for this repository can be applied. See apt-secure(8) manpage for details.
Do you want to accept these changes and continue updating from this repository? [y/N]
Do you want to accept these changes and continue updating from this repository? [y/N]
L@Dr. Furkan Gözükara so, what's the BEST dreambooth model you've ever trained?
LI'm looking at this post: https://www.patreon.com/posts/best-stable-1-5-90307694 - but did you only train it for 4 epochs and 1,000 steps?
Patreon
Get more from SECourses: Tutorials, Guides, Resources, Training, MidJourney, Voice Clone, TTS, ChatGPT, GPT, LLM, Scripts by Furkan Gözükara on Patreon
Fthis is in research
Fmy best is sdxl dreambooth
Fcan you private message me with screenshot?
Fi will update notebook which one and error is where
L
Ltyty! Will check it out and report back
Linteresting - surprised euler_a is the best sample sampler
Lalso, 1000 steps for 4 epochs right? How many images did you use and how long did the whole training test take btw?
LAlso, does choosing 16 24 or 48GB affect the output? or just the speed? What is the tradeoff
