Diffusion Transformer explained (Stable Diffusion 3)
Ғылым және технология
The concept of Diffusion Transformer explained. Plus the forward and backward diffusion process and the critical dimensionality of the latent space. Detailed explanation how noise is added from a Gaussian distribution and outlook on further devs for Diffusion based Text-to-Image models. Plus: Rectified Flow as a new tech implementation in Stable Diffusion 3.
#newtechnology
#airesearch
#ai
Пікірлер: 4
I'm glad you took the time to explain how the word "rectified" is used in this context. I was confused at first, because rectification has a pretty specific meaning in other domains that doesn't logically pertain to this discussion. I'm really curious about this symmetry-breaking exploit you mention. Great lectures!
For 1500€ you can get 2 second hand rtx3090 and nvlink running in sli, a matx and a pcie raiser to fit them in the case for less than 200€, the power of the system might be the higher one but if you need it i think its afordable. Also you could run proxmox and virtualize the gpus for different tasks or run them together for 48GB of VRAM and 1TB/s memory bandwidth per card
Thank you for the update!!!