Dimensionality of `LanguageBind_Video_Huge_V1.5_FT`

[pizhn]

Hi,

Upon checking the large video model LanguageBind_Video_Huge_V1.5_FT, I noticed that its embedding dimension is 1024, whereas the others use 768. However, I couldn’t find a corresponding language backbone with the same embedding size. Could you clarify which language model should be used to match this embedding space? Thank you.

  (modality_proj): ModuleDict(
    (video): Linear(in_features=1280, out_features=1024, bias=False)
    (image): Linear(in_features=1024, out_features=768, bias=False)
    (language): Linear(in_features=768, out_features=768, bias=False)
  )