Recorded this one one go, messy video, might make another late. #dreambooth #stablediffusion #nvidia #aitraining #advice #helpful #interesting
Пікірлер: 25
@JimPoisonFishingPlanet Жыл бұрын
If you have low GPU RAM available, make sure to add a pipe.enable_attention_slicing() after sending it to cuda for less VRAM usage (to the cost of speed) So the question is, where to enter this code? Straight to promt? I have 4 gigabytes of RAM, which only did not SD 2 version swears that there is not enough memory CUDA
@w2lf Жыл бұрын
After creating the classifiers it runs out of memory.
@thevaldis1167 Жыл бұрын
Does it work with GTX 1050 Ti (4gb VRAM)? I tried earlier and I got error "CUDA Run Out Of Memory" right before finalizing training
@miguelarce6489 Жыл бұрын
What vram you used here?
@victorproff3481 Жыл бұрын
I have RTX3050 , 4Gb VRam, will it work?
@sukhpalsukh3511 Жыл бұрын
I have 3060ti 8gb vram, using kohya_ss ui with same settings xformers , but training gets error , tried Lora training and it's running , but checkpoint training nit working.
@sukhpalsukh3511 Жыл бұрын
Could you make a tutorial on 8gb vram, and newer version of kohya_ss gui with dreambooth, there's one more option in for lowvram , like deepspeed option, but unable to understand that ,
@jorgetinoco8249 Жыл бұрын
I have an RTX 3070 (8GB), but even with it I couldn't get past the first step creating the model (Create Model) it says in the webui: "Error", and in the console it says: "IndexError: tuple index out of range ", does anyone know how fix it?.
@archerthephantom6942 Жыл бұрын
which version of stable diffusion are you running? 1.4, 1.5 or 2? Also which model have you selected to train?
@Hexagon666 Жыл бұрын
I got the same GPU and I also got this error. However I only get this when selecting "waifu diffusion v1.3" but "Stable diffusion v1.4", "Waifu Diffusion v1.2" , the default "model.ckpt" etc, worked fine. I noticed that waifu diffusion v1.3 has "opt" in its filename while the others don't. So maybe opt models won't work? But in the end when I want to start training, everything works until the actual training starts. Then I hear some coil whine from my GPU and I get an error message saying: RuntimeError: CUDA out of memory. Tried to allocate 10.00 MiB (GPU 0; 8.00 GiB total capacity; 7.16 GiB already allocated; 0 bytes free; 7.27 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
@jorgetinoco8249 Жыл бұрын
@@archerthephantom6942 For both the 1.5 version. Thank you.
@archerthephantom6942 Жыл бұрын
@@jorgetinoco8249 There are 2 ckpt files for 1.5, one is around 7 gb other is the pruned 4-5 gb one, try using the smaller one if you aren't already. I remember getting the same error but had no idea what I did to fix it, you can try it with xformers if you have not already. ( edit your webui.bat with a notepad and put --xformers to the command arguments). Also one thing I would suggest it using google colab instead of training locally if you can, honestly it is so much nicer and you can train full prevision there.
@archerthephantom6942 Жыл бұрын
@@Hexagon666 you have to make sure you have --xformers in your webui.bat, xformers selected in advanced and fp16 selected. Also Select 8bitAdams as well
@eliasherreradg5966 Жыл бұрын
i have a GPU amd , but it gives me an error when creating the model.
@archerthephantom6942 Жыл бұрын
I don't think you can use an amd gpu for training. You can try to do it with google colab instead.
@rilijohnmichael Жыл бұрын
In my memory attention option, I only have "flash_attention" option and there is no xformers option. Running 1.5 AND DREAMBOOTH IS ON LATEST VERSION.
@archerthephantom6942 Жыл бұрын
Make sure you have xformers installed before launching webui.bat, you can do this by opening webui.bat with a notepad and adding --xformers to the command arguments section. End result should look like this: set COMMANDLINE_ARGS= --xformers
@Igor-rq5mp Жыл бұрын
Hey, i did exactly like you and got this error "Please configure some concepts." do you know why?
@archerthephantom6942 Жыл бұрын
Hey hope you found a sloution already, if not can try to help you out.
@Igor-rq5mp Жыл бұрын
@@archerthephantom6942 i didnt :(
@archerthephantom6942 Жыл бұрын
@@Igor-rq5mp did you get this error when you started the training? Did you fill the concepts tab properly? In this video I basicly fill the concept tab with useless and improper stuff to get to training as soon as possible so hopefully you did fill it properly.
@Igor-rq5mp Жыл бұрын
@@archerthephantom6942 yes, when i click "traing" i get this error I did exactly like you did, at least it should have trained, even though it was gibberish in the concepts tab i also did like some others tutorials and got the exact same result, i will try to reinstall the a1111 and see what happens
@katze_ksb Жыл бұрын
@@Igor-rq5mp when creating a model don't click save params or load params, because those buttons are bugged. Just klick "train" and react to the dialog window that will pop up.
@mlnima Жыл бұрын
I am thinking to sell my kidny and but A6000 with 48gb :D, no just kidding