What is the difference between the A14B and 14B models?
Hello Kijai, I am not sure if this the right forum to ask this, but could you or somebody else explain to me the difference between the A14B and 14B models? What does the 'A' stand for and how exactly is it different? I can't find an explanation anywhere
I liked to know that as well. I remember googling it and I feel the AI wasn't giving me the right answer. Also, what was exactly was fixed in Wan2.2-Lightning_T2V-v1.1-A14B? Over Wan2.2-Lightning_T2V-A14B? Could I get some 1.1 patch notes?
Doesn't really matter. I just go with the latest version. I get amazing results nonetheless.
I've asked in the lightx2v repository too....no response. We will never know :D
As far as I know the "A" refers to "active" in the MoE (mixture of experts) architecture, that this kind of model setup is.
Meaning we have different models for different part of the same process: the high noise model to start and low noise model to finish the generation.