About the learnable skip token for SPHINX_X

[jungle-gym-ac]

Based on the code I've read in Here, the Learnable Skip Token mentioned in the Sphinx-X paper has not been implemented in the SPHINX-v2 version. The improvements over the original SPHINX code include removing CLIP-ViT and Q-Former, and changing the input resolution of ConvNeXT to 1024.

May I ask when the code corresponding to the skip token will be open-sourced?