3 ms·
> It’s crucial to highlight the resource efficiency of our training process. Our training is considerably more resource-efficient compared to Stable Diffusion (
by littlestymaar 2y ago
> It’s crucial to highlight the resource efficiency of our training process. Our training is considerably more resource-efficient compared to Stable Diffusion (Podell et al., 2023). Meissonic is trained in approximately 48 H100 GPU days
From scratch training of an image synthesis model for the price of a graphic card isn't something I expected anytime soon!