beginner-level question on conditioning

[BLT1967]

Do you have any advice on where in this code to attach a small (e.g., 4-dimensional or 8-dimensional) conditioning vector? Or if it cannot be done?

[Francesco215]

Hi,

you may change it by modifying this line here to enrich the way conditioning is applied

text-diffusion/continous_diffusion/model.py

Line 36 in 628b1c8

conditioning = self.time_conditioning(c_noise)

[BLT1967]

Hi, thanks for responding so promptly! I'm sure this will seem like a totally trivial question, but I'm still finding my way around the code base (and text diffusion in general); can I simply concatenate a conditioning vector? Won't this affect the dimensioning? Or is there a more elegant way of doing this?

(Also, are you responding from Italy?)

[Francesco215]

You'll have to change a bit the code. For example you can pass the conditioning vector and then at each layer you project it to the shape you need

In general I would advice you against doing serious experiments with CDCD because it's not really scalable.

take a look at this instead https://arxiv.org/pdf/2502.09992

P.S.
Yep, I'm in Italy at this moment