exo
exo copied to clipboard
video inference
Could this system be used to accumulate enough RAM to run a video model like Hunyuan Video?
Yes, it looks like a good fit for exo since the architecture is parallelisable and requires a lot of memory.
Are there examples of how to set up something with such a huge amount of VRAM? I don't know if it would be cheaper to combine a bunch of small phones, Raspberry Pis, or larger PCs. I have 64GB of RAM, but the model doesn't let me use it in CPU+RAM; it forces me to use CUDA and VRAM.