memory problem

[hiop]

Hello. How much memory do you need to run?

RuntimeError: CUDA out of memory. Tried to allocate 20.00 MiB (GPU 0; 6.00 GiB total capacity; 4.78 GiB already allocated; 0 bytes free; 4.82 GiB reserved in total by PyTorch)

Any solution to run with a small amount of memory?

[vipermu]

some stuff that can help (from https://t.co/KctR4TMotT)

• in text2img.py, line 18 change map_location="cpu" for map_location="cuda"
• you can remove line 29 and add model.half()
• then, add the following between line 119 and 120 with torch.cuda.amp.autocast():

[hiop]

@vipermu The process has begun, although the result is green squares

[emilonpt]

I'm unable to run this on a 6gb graphics card due to the same memory error, even with the changes proposed.

Is there anything else I could try or a way to run a smaller mode?

[serhio-k]

I'm unable to run this on a 6gb graphics card due to the same memory error, even with the changes proposed.

Is there anything else I could try or a way to run a smaller mode?

Unfortunately the more or less appropriate results can be achived with 8gb or more VRAM. Nevertheless you can try to reduce the number of pictures by changing --n_samples and --n_iter values. You could also change the output resolution for the images. For example try this: python scripts/txt2img.py --prompt "a dog eating a hamburger in space" --ddim_eta 0.0 --n_samples 3 --n_iter 3 --scale 5.0 --ddim_steps 100 --H 192 --W 192

[adelin-b]

same here, the process get killed right after the start.

python scripts/txt2img.py --prompt "a dog eating a hamburger in space" --ddim_eta 0.0 --n_samples 3 --n_iter 3 --scale 5.0 --ddim_steps 100 --H 192 --W 192
Loading model from models/ldm/text2img-large/model.ckpt
LatentDiffusion: Running in eps-prediction mode
DiffusionWrapper has 872.30 M params.
making attention of type 'vanilla' with 512 in_channels
Working with z of shape (1, 4, 32, 32) = 4096 dimensions.
making attention of type 'vanilla' with 512 in_channels
[1]    12177 killed     python scripts/txt2img.py --prompt "a dog eating a hamburger in space"  0.0 

The machine itself

CPU: 11th Gen Intel i9-11900H (16) @ 4.800GHz 
GPU: Intel TigerLake-H GT1 [UHD Graphics] 
GPU: NVIDIA GeForce RTX 3060 Mobile / Max-Q 
Memory: 5752MiB / 15678MiB (36%) 

I wonder if its because the script is trying to use my intel graphic card instead of nvidia

[wes-kay]

@adelin-b I'm thinking the same thing, did you find a solution?

[wes-kay]

@adelin-b Looks like with os.environ["CUDA_VISIBLE_DEVICES"]=<YOUR_GPU_NUMBER_HERE> my laptop is using the correct gpu

[kalpa61]

@hiop hi,Have you solved the problem of running out of memory

[hiop]

@hiop hi,Have you solved the problem of running out of memory

yes I bought 4070

[kalpa61]

@hiop ok,thank you