internvideo2 distilled models config

[orrzohar]

Hi,
For the InternVideo2-S/B/L encoders: what value was used for sep_image_video_pos_embed? It seems like this was set true in the 1b/6b models, but false in S/B/L

I am trying to run inference but getting very poor results.

Best,
Orr