search

Vchitect / Latte

Latte: Latent Diffusion Transformer for Video Generation.
Apache License 2.0
1.67k stars 177 forks source link

How can I utilize the weights of pre-trained PixArt-α to initialize the parameters of the spatial Transformer block in the Latte T2V model? #95

Closed g-makerr closed 3 months ago

g-makerr commented 3 months ago

How can I utilize the weights of pre-trained PixArt-α to initialize the parameters of the spatial Transformer block in the Latte T2V model.

maxin-cn commented 3 months ago

You can use the weights of pre-trained PixArt-α to initialize this part while making the other part random initialization.

g-makerr commented 3 months ago

You can use the weights of pre-trained PixArt-α to initialize this part while making the other part random initialization.

Thanks for your reply. I will try it.