I made a colab demo but i get 2 erros

[francqz31]

Hi Authors , Here is a full demo Colab that you can build on , but I get these 2 errors

https://colab.research.google.com/drive/13BIlDBZK4rcrzLxK3N7tamrPcvX6lAIB#scrollTo=HcBUTwmFjpUT

whenever i run the inference or the training scripts I get this error subprocess.CalledProcessError: Command '['/usr/local/bin/python3.10', 'train_t2i_custom_v2.py', '--config=configs/custom.py']' died with <Signals.SIGKILL: 9>.

and whenever i run this !python gradio_demo.py i get this error 2023-07-04 15:37:43.188 | DEBUG | open_clip.transformer:init:314 - xattn in transformer of CLIP is False 2023-07-04 15:38:02.273 | DEBUG | open_clip.transformer:init:314 - xattn in transformer of CLIP is False ^C I'm not sure what's happening but if you can identify the problem that would be amazing. Thanks in advance

[loboere]

could you fix this?

[francqz31]

not really , it is either a problem with colab's environment, or a problem in the existing code. I can't try it on my pc for now. so I have to wait for a response.

[loboere]

the first time i run !python gradio_demo.py it seems to download a 10.2gb file

Downloading (…)ip_pytorch_model.bin: 94% 9.56G/10.2G [09:16<00:22, 26.9MB/s

then I guess it tries to load into memory but with the memory that colab gives free it is not enough so it gives that error. It would be a shame if this is it.

You would have to try colab pro if you have colab pro you can try it.

[zideliu]

The gradio demo is now deployed to HuggingFace space, available here. Unlike gradio_demo in this repo, we useprompt_model,_,_ = open_clip.create_model_and_transforms('ViT-bigG-14', 'laion2b_s39b_b160k',device='cpu') which use 'fp16' as default

[ship-it-mind]

@loboere @francqz31 I faced the same error on colab the free version but after upgrading to the pro version it worked. So the issue happens because of not enough RAM or VRAM