Open alta750 opened 1 year ago
maybe is because your GPU has only 4GB of VRAM, what image size was your prompt made with?
I was trying with 512 x 512
i seem to be having the same problem with a 10gb 3080 card. even if i just boot linux right into cmd with zero usage of gfx ram i still have it complaining about memory. and that is with a 512 x 512 pic...
I tried this instead: https://github.com/AUTOMATIC1111/stable-diffusion-webui
with --medvram --opt-split-attention
in the webui-user.bat
file, which works for me just fine. You might not need that command though, since you have 10gb vram
Line 62, replace model.cuda()
by model.cuda().half()
for me the parameters --medvram and --opt-split-attention are not available on txt2imghd.py - did i miss a step during "installation"? i tried recreating it from copying parts of the optimized/webui but without success. Any ideas on how to get it running on low VRAM GPUs yet?
for me the parameters --medvram and --opt-split-attention are not available on txt2imghd.py - did i miss a step during "installation"? i tried recreating it from copying parts of the optimized/webui but without success. Any ideas on how to get it running on low VRAM GPUs yet?
Those parameters are for a different one, you need to download this first.
Follow the install instruction and then in the webui-user.bat
file you'll put --medvram --opt-split-attention
in the commandline_args
I've added
model.half()
underneathmodel = instantiate_from_config(config.model)
andinit_image = init_image.half()
underneathinit_image = repeat(init_image, '1 ... -> b ...', b=batch_size)
I get this error:
RuntimeError: CUDA out of memory. Tried to allocate 1024.00 MiB (GPU 0; 4.00 GiB total capacity; 2.56 GiB already allocated; 183.30 MiB free; 2.58 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
What should I do?