search

lucidrains / deep-daze

Simple command line tool for text to image generation using OpenAI's CLIP and Siren (Implicit neural representation network). Technique was originally created by https://twitter.com/advadnoun
MIT License
4.37k stars 327 forks source link

"Killed" when imagining #105

Open dzoba opened 3 years ago

dzoba commented 3 years ago 
(myenv) chris@chris-desktop:~$ imagine "the most beautiful image in the world"
Starting up...
Imagining "the_most_beautiful_image_in_the_world" from the depths of my weights...
Killed

I'm seeing this message when I try to imagine. I am running this on an Nvidia Jetson Nano (the 4gb version).

These are the versions I am running

(myenv) chris@chris-desktop:~$ python --version
Python 3.6.9
(myenv) chris@chris-desktop:~$ python
Python 3.6.9 (default, Jan 26 2021, 15:33:00) 
[GCC 8.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> print(torch.__version__)
1.8.0
>>> import torchvision
>>> print(torchvision.__version__)
0.9.0
dzoba commented 3 years ago

I was able to get beyond this error by increasing the swap memory on the Jetson following these directions: https://github.com/JetsonHacksNano/resizeSwapMemory

However, now when I begin to imagine I get cuda error: the launch timed out and was terminated

NotNANtoN commented 3 years ago

I would assume that the process was killed due to a lack in RAM. What you can try is to reduce the image_width and batch_sizeto an acceptable level.

I have no clue about the cuda launch time-out error.

DannyBen commented 2 years ago

Where can we learn about the minimum RAM requirements?