Full and dev is using too much vram
#4
by
Sumitc13
- opened
While fast move is working fine but the full and dev modes are using too much vram, more than 25GB.
Same here. My Jetson AGX Orin with 32GB (Ampere gpu) can run the fast version (it takes over a minute to generate one image, which is very slow compared to other image generators). There is no swap needed for FAST.
The DEV version took longer to generate an image, but also filled up my swap for about 14GB or so.
I also see messages telling me 'took more than 77 tokens .. ..', how short do we need to make the prompt, and why would we then even use llama?